Skip to contentSkip to conversation

Configuring traffic classes via new hardware offload mechanism in tc/mqprio

PATCH Superseded Amritha Nambiar 3 participants 14 messages Aug 17, 2017 – Aug 18, 2017

Subsystems (4)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover Configuring traffic classes via new hardware offload mechanism in tc/mqprio

The following series introduces a new hardware offload mode in tc/mqprio where the TCs, the queue configurations and bandwidth rate limits are offloaded to the hardware. The existing mqprio framework is extended to configure the queue counts and layout and also added support for rate limiting. This is achieved by setting the value 2 for the "hw" mode. Legacy devices can fall back to the existing setup supporting hw mode 1 where only the TCs are offloaded, however if hw mode 2 is specified, then this type of offload fails to initialize if the requested mode is not supported. The i40e driver enables the new mqprio hardware offload mechanism factoring the TCs, queue configuration and bandwidth rates by creating HW channel VSIs.

In this new mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic class when the 'hw' option is set to 2. This is achieved by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0) which is for the main VSI. TC0 for the main VSI is also reconfigured as per user provided queue parameters. Finally, bandwidth rate limits are set on these traffic classes through the mqprio offload framework by sending these rates in addition to the number of TCs and the queue configurations.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 2  map 0 0 0 0 1 1 1 1\
  queues 4 at 0 4 at 4 max_rate 55Mbit 60Mbit hw 2

To dump the bandwidth rates:

# tc qdisc show dev eth0
  qdisc mqprio 804a: root  tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
               queues:(0:3) (4:7)
               min_rate:0bit 0bit
               max_rate:55Mbit 60Mbit

---

Amritha Nambiar (6):
[net-next]net: mqprio: Introduce new hardware offload mode in mqprio for offloading full TC configurations [net-next]net: i40e: Add macro for PF reset bit
[net-next]net: i40e: Add infrastructure for queue channel support with the TCs and queue configurations offloaded via mqprio scheduler [net-next]net: i40e: Enable mqprio full offload mode in the i40e driver for configuring TCs and queue mapping [net-next]net: i40e: Refactor VF BW rate limiting function to be reused on the PF as well
[net-next]net: i40e: Add support to set max bandwidth rates for TCs offloaded via tc/mqprio

 drivers/net/ethernet/intel/i40e/i40e.h             |   45 +
 drivers/net/ethernet/intel/i40e/i40e_debugfs.c     |    3 
 drivers/net/ethernet/intel/i40e/i40e_ethtool.c     |    8 
 drivers/net/ethernet/intel/i40e/i40e_main.c        | 1520 +++++++++++++++++---
 drivers/net/ethernet/intel/i40e/i40e_txrx.h        |    2 
 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c |   50 -
 include/linux/netdevice.h                          |    1 
 include/net/pkt_cls.h                              |    7 
 include/uapi/linux/pkt_sched.h                     |   13 
 net/sched/sch_mqprio.c                             |  167 ++
 10 files changed, 1563 insertions(+), 253 deletions(-)

--

#1 lore raw

PATCH v2 1/6 net: mqprio: Introduce new hardware offload mode in mqprio for offloading full TC configurations

This patch introduces a new hardware offload mode in mqprio which makes full use of the mqprio options, the TCs, the queue configurations and the bandwidth rates for the TCs. This is achieved by setting the value 2 for the "hw" option. This new offload mode supports new attributes for traffic class such as minimum and maximum values for bandwidth rate limits.

Introduces a new datastructure 'tc_mqprio_qopt_offload' for offloading mqprio queue options and use this to be shared between the kernel and device driver. This contains a copy of the exisiting datastructure for mqprio queue options. This new datastructure can be extended when adding new attributes for traffic class such as bandwidth rate limits. The existing datastructure for mqprio queue options will be shared between the kernel and userspace.

This patch enables configuring additional attributes associated with a traffic class such as minimum and maximum bandwidth rates and can be offloaded to the hardware in the new offload mode. The min and max limits for bandwidth rates are provided by the user along with the the TCs and the queue configurations when creating the mqprio qdisc.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\
  queues 4 at 0 4 at 4 min_rate 10Mbit 20Mbit max_rate 55Mbit 60Mbit hw 2

To dump the bandwidth rates:

# tc qdisc show dev eth0

qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0

             queues:(0:3) (4:7)
             min_rate:10Mbit 20Mbit
             max_rate:55Mbit 60Mbit

v2 : Jiri's changes accepted upstream removes the struct tc_to_netdev. Clean up the full offload related changes added to mqprio_init() and mqprio_destroy() to rebase on these changes.

---
 include/linux/netdevice.h      |    1 
 include/net/pkt_cls.h          |    7 ++
 include/uapi/linux/pkt_sched.h |   13 +++
 net/sched/sch_mqprio.c         |  167 ++++++++++++++++++++++++++++++++++++++--
 4 files changed, 178 insertions(+), 10 deletions(-)
include/linux/netdevice.h +1 -0
index 0f1c4cb..631f314 100644
@@ -777,6 +777,7 @@ enum tc_setup_type {
 	TC_SETUP_CLSFLOWER,
 	TC_SETUP_CLSMATCHALL,
 	TC_SETUP_CLSBPF,
+	TC_SETUP_MQPRIO_EXT,
 };
 
 /* These structures hold the attributes of xdp state that are being passed
include/net/pkt_cls.h +7 -0
index e80edd8..55bf148 100644
@@ -546,6 +546,13 @@ struct tc_cls_bpf_offload {
 	u32 gen_flags;
 };
 
+struct tc_mqprio_qopt_offload {
+	/* struct tc_mqprio_qopt must always be the first element */
+	struct tc_mqprio_qopt qopt;
+	u32 flags;
+	u64 min_rate[TC_QOPT_MAX_QUEUE];
+	u64 max_rate[TC_QOPT_MAX_QUEUE];
+};
 
 /* This structure holds cookie structure that is passed from user
  * to the kernel for actions and classifiers
include/uapi/linux/pkt_sched.h +13 -0
index 099bf55..cf2a146 100644
@@ -620,6 +620,7 @@ struct tc_drr_stats {
 enum {
 	TC_MQPRIO_HW_OFFLOAD_NONE,	/* no offload requested */
 	TC_MQPRIO_HW_OFFLOAD_TCS,	/* offload TCs, no queue counts */
+	TC_MQPRIO_HW_OFFLOAD,		/* fully supported offload */
 	__TC_MQPRIO_HW_OFFLOAD_MAX
 };
 
@@ -633,6 +634,18 @@ struct tc_mqprio_qopt {
 	__u16	offset[TC_QOPT_MAX_QUEUE];
 };
 
+#define TC_MQPRIO_F_MIN_RATE  0x1
+#define TC_MQPRIO_F_MAX_RATE  0x2
+
+enum {
+	TCA_MQPRIO_UNSPEC,
+	TCA_MQPRIO_MIN_RATE64,
+	TCA_MQPRIO_MAX_RATE64,
+	__TCA_MQPRIO_MAX,
+};
+
+#define TCA_MQPRIO_MAX (__TCA_MQPRIO_MAX - 1)
+
 /* SFB */
 
 enum {
net/sched/sch_mqprio.c +157 -10

231 lines not on this page · raw message

#2 lore raw

PATCH v2 1/6 net: mqprio: Introduce new hardware offload mode in mqprio for offloading full TC configurations

Two little thoughts...
sln

Quoted 113 lines · Amritha Nambiar

This patch introduces a new hardware offload mode in mqprio which makes full use of the mqprio options, the TCs, the queue configurations and the bandwidth rates for the TCs. This is achieved by setting the value 2 for the "hw" option. This new offload mode supports new attributes for traffic class such as minimum and maximum values for bandwidth rate limits.

Introduces a new datastructure 'tc_mqprio_qopt_offload' for offloading mqprio queue options and use this to be shared between the kernel and device driver. This contains a copy of the exisiting datastructure for mqprio queue options. This new datastructure can be extended when adding new attributes for traffic class such as bandwidth rate limits. The existing datastructure for mqprio queue options will be shared between the kernel and userspace.

This patch enables configuring additional attributes associated with a traffic class such as minimum and maximum bandwidth rates and can be offloaded to the hardware in the new offload mode. The min and max limits for bandwidth rates are provided by the user along with the the TCs and the queue configurations when creating the mqprio qdisc.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\
   queues 4 at 0 4 at 4 min_rate 10Mbit 20Mbit max_rate 55Mbit 60Mbit hw 2

To dump the bandwidth rates:

# tc qdisc show dev eth0

qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0

              queues:(0:3) (4:7)
              min_rate:10Mbit 20Mbit
              max_rate:55Mbit 60Mbit

v2 : Jiri's changes accepted upstream removes the struct tc_to_netdev. Clean up the full offload related changes added to mqprio_init() and mqprio_destroy() to rebase on these changes.

---
  include/linux/netdevice.h      |    1
  include/net/pkt_cls.h          |    7 ++
  include/uapi/linux/pkt_sched.h |   13 +++
  net/sched/sch_mqprio.c         |  167 ++++++++++++++++++++++++++++++++++++++--
  4 files changed, 178 insertions(+), 10 deletions(-)
include/linux/netdevice.h +1 -0
index 0f1c4cb..631f314 100644
@@ -777,6 +777,7 @@ enum tc_setup_type {
  	TC_SETUP_CLSFLOWER,
  	TC_SETUP_CLSMATCHALL,
  	TC_SETUP_CLSBPF,
+	TC_SETUP_MQPRIO_EXT,
  };
 
  /* These structures hold the attributes of xdp state that are being passed
include/net/pkt_cls.h +7 -0
index e80edd8..55bf148 100644
@@ -546,6 +546,13 @@ struct tc_cls_bpf_offload {
  	u32 gen_flags;
  };
 
+struct tc_mqprio_qopt_offload {
+	/* struct tc_mqprio_qopt must always be the first element */
+	struct tc_mqprio_qopt qopt;
+	u32 flags;
+	u64 min_rate[TC_QOPT_MAX_QUEUE];
+	u64 max_rate[TC_QOPT_MAX_QUEUE];
+};
 
  /* This structure holds cookie structure that is passed from user
   * to the kernel for actions and classifiers
include/uapi/linux/pkt_sched.h +13 -0
index 099bf55..cf2a146 100644
@@ -620,6 +620,7 @@ struct tc_drr_stats {
  enum {
  	TC_MQPRIO_HW_OFFLOAD_NONE,	/* no offload requested */
  	TC_MQPRIO_HW_OFFLOAD_TCS,	/* offload TCs, no queue counts */
+	TC_MQPRIO_HW_OFFLOAD,		/* fully supported offload */
  	__TC_MQPRIO_HW_OFFLOAD_MAX
  };
 
@@ -633,6 +634,18 @@ struct tc_mqprio_qopt {
  	__u16	offset[TC_QOPT_MAX_QUEUE];
  };
 
+#define TC_MQPRIO_F_MIN_RATE  0x1
+#define TC_MQPRIO_F_MAX_RATE  0x2
+
+enum {
+	TCA_MQPRIO_UNSPEC,
+	TCA_MQPRIO_MIN_RATE64,
+	TCA_MQPRIO_MAX_RATE64,
+	__TCA_MQPRIO_MAX,
+};
+
+#define TCA_MQPRIO_MAX (__TCA_MQPRIO_MAX - 1)
+
  /* SFB */
 
  enum {
net/sched/sch_mqprio.c +0 -0
index 2165a05..d913111 100644

net/sched/sch_mqprio.c line 27

Quoted 10 lines
net/sched/sch_mqprio.c +3 -0
@@ -18,10 +18,13 @@
  #include <net/netlink.h>
  #include <net/pkt_sched.h>
  #include <net/sch_generic.h>
+#include <net/pkt_cls.h>
 
  struct mqprio_sched {
  	struct Qdisc		**qdiscs;
  	int hw_offload;
+	u32 flags;
+	u64 min_rate[TC_QOPT_MAX_QUEUE], max_rate[TC_QOPT_MAX_QUEUE];

Put these on separate lines.

Quoted 115 lines
net/sched/sch_mqprio.c +80 -2
  };
 
  static void mqprio_destroy(struct Qdisc *sch)
@@ -39,9 +42,20 @@ static void mqprio_destroy(struct Qdisc *sch)
  	}
 
  	if (priv->hw_offload && dev->netdev_ops->ndo_setup_tc) {
-		struct tc_mqprio_qopt mqprio = {};
+		struct tc_mqprio_qopt_offload mqprio = { { 0 } };
 
-		dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO, &mqprio);
+		switch (priv->hw_offload) {
+		case TC_MQPRIO_HW_OFFLOAD_TCS:
+			dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO,
+						      &mqprio.qopt);
+			break;
+		case TC_MQPRIO_HW_OFFLOAD:
+			dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO_EXT,
+						      &mqprio);
+			break;
+		default:
+			return;
+		}
  	} else {
  		netdev_set_num_tc(dev, 0);
  	}
@@ -97,6 +111,24 @@ static int mqprio_parse_opt(struct net_device *dev, struct tc_mqprio_qopt *qopt)
  	return 0;
  }
 
+static const struct nla_policy mqprio_policy[TCA_MQPRIO_MAX + 1] = {
+	[TCA_MQPRIO_MIN_RATE64] = { .type = NLA_NESTED },
+	[TCA_MQPRIO_MAX_RATE64] = { .type = NLA_NESTED },
+};
+
+static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
+		      const struct nla_policy *policy, int len)
+{
+	int nested_len = nla_len(nla) - NLA_ALIGN(len);
+
+	if (nested_len >= nla_attr_size(0))
+		return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
+				 nested_len, policy, NULL);
+
+	memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
+	return 0;
+}
+
  static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  {
  	struct net_device *dev = qdisc_dev(sch);
@@ -105,6 +137,10 @@ static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  	struct Qdisc *qdisc;
  	int i, err = -EOPNOTSUPP;
  	struct tc_mqprio_qopt *qopt = NULL;
+	struct nlattr *tb[TCA_MQPRIO_MAX + 1];
+	struct nlattr *attr;
+	int rem;
+	int len = nla_len(opt) - NLA_ALIGN(sizeof(*qopt));
 
  	BUILD_BUG_ON(TC_MAX_QUEUE != TC_QOPT_MAX_QUEUE);
  	BUILD_BUG_ON(TC_BITMASK != TC_QOPT_BITMASK);
@@ -122,6 +158,51 @@ static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  	if (mqprio_parse_opt(dev, qopt))
  		return -EINVAL;
 
+	if (len > 0) {
+		err = parse_attr(tb, TCA_MQPRIO_MAX, opt, mqprio_policy,
+				 sizeof(*qopt));
+		if (err < 0)
+			return err;
+
+		if (tb[TCA_MQPRIO_MIN_RATE64]) {
+			if (qopt->hw != TC_MQPRIO_HW_OFFLOAD)
+				return -EINVAL;
+
+			i = 0;
+			nla_for_each_nested(attr, tb[TCA_MQPRIO_MIN_RATE64],
+					    rem) {
+				if (nla_type(attr) != TCA_MQPRIO_MIN_RATE64)
+					return -EINVAL;
+
+				if (i >= qopt->num_tc)
+					return -EINVAL;
+
+				priv->min_rate[i] = *(u64 *)nla_data(attr);
+				i++;
+			}
+			priv->flags |= TC_MQPRIO_F_MIN_RATE;
+		}
+
+		if (tb[TCA_MQPRIO_MAX_RATE64]) {
+			if (qopt->hw != TC_MQPRIO_HW_OFFLOAD)
+				return -EINVAL;
+
+			i = 0;
+			nla_for_each_nested(attr, tb[TCA_MQPRIO_MAX_RATE64],
+					    rem) {
+				if (nla_type(attr) != TCA_MQPRIO_MAX_RATE64)
+					return -EINVAL;
+
+				if (i >= qopt->num_tc)
+					return -EINVAL;
+
+				priv->max_rate[i] = *(u64 *)nla_data(attr);
+				i++;
+			}
+			priv->flags |= TC_MQPRIO_F_MAX_RATE;
+		}
+	}
+
  	/* pre-allocate qdisc, attachment can't fail */
  	priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
  			       GFP_KERNEL);

net/sched/sch_mqprio.c static int mqprio_init(struct Qdisc *sch, struct nlattr *opt) · line 229

Quoted 3 lines
net/sched/sch_mqprio.c +0 -0
@@ -146,14 +227,35 @@ static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  	 * supplied and verified mapping
  	 */
  	if (qopt->hw) {

In the changes above is a test for existence of ->ndo_setup_tc(), should there be a similar test here as well?

Quoted 116 lines
net/sched/sch_mqprio.c +74 -8
-		struct tc_mqprio_qopt mqprio = *qopt;
+		struct tc_mqprio_qopt_offload mqprio = {.qopt = *qopt};
 
-		err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO,
-						    &mqprio);
+		switch (qopt->hw) {
+		case TC_MQPRIO_HW_OFFLOAD_TCS:
+			err = dev->netdev_ops->ndo_setup_tc(dev,
+							    TC_SETUP_MQPRIO,
+							    &mqprio.qopt);
+			break;
+		case TC_MQPRIO_HW_OFFLOAD:
+			mqprio.flags = priv->flags;
+			if (priv->flags & TC_MQPRIO_F_MIN_RATE)
+				for (i = 0; i < mqprio.qopt.num_tc; i++)
+					mqprio.min_rate[i] = priv->min_rate[i];
+
+			if (priv->flags & TC_MQPRIO_F_MAX_RATE)
+				for (i = 0; i < mqprio.qopt.num_tc; i++)
+					mqprio.max_rate[i] = priv->max_rate[i];
+
+			err = dev->netdev_ops->ndo_setup_tc(dev,
+							    TC_SETUP_MQPRIO_EXT,
+							    &mqprio);
+			break;
+		default:
+			return -EINVAL;
+		}
  		if (err)
  			return err;
 
-		priv->hw_offload = mqprio.hw;
+		priv->hw_offload = mqprio.qopt.hw;
  	} else {
  		netdev_set_num_tc(dev, qopt->num_tc);
  		for (i = 0; i < qopt->num_tc; i++)
@@ -223,11 +325,51 @@ static int mqprio_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
  	return 0;
  }
 
+static int dump_rates(struct mqprio_sched *priv,
+		      struct tc_mqprio_qopt *opt, struct sk_buff *skb)
+{
+	struct nlattr *nest;
+	int i;
+
+	if (priv->flags & TC_MQPRIO_F_MIN_RATE) {
+		nest = nla_nest_start(skb, TCA_MQPRIO_MIN_RATE64);
+		if (!nest)
+			goto nla_put_failure;
+
+		for (i = 0; i < opt->num_tc; i++) {
+			if (nla_put(skb, TCA_MQPRIO_MIN_RATE64,
+				    sizeof(priv->min_rate[i]),
+				    &priv->min_rate[i]))
+				goto nla_put_failure;
+		}
+		nla_nest_end(skb, nest);
+	}
+
+	if (priv->flags & TC_MQPRIO_F_MAX_RATE) {
+		nest = nla_nest_start(skb, TCA_MQPRIO_MAX_RATE64);
+		if (!nest)
+			goto nla_put_failure;
+
+		for (i = 0; i < opt->num_tc; i++) {
+			if (nla_put(skb, TCA_MQPRIO_MAX_RATE64,
+				    sizeof(priv->max_rate[i]),
+				    &priv->max_rate[i]))
+				goto nla_put_failure;
+		}
+		nla_nest_end(skb, nest);
+	}
+	return 0;
+
+nla_put_failure:
+	nla_nest_cancel(skb, nest);
+	return -1;
+}
+
  static int mqprio_dump(struct Qdisc *sch, struct sk_buff *skb)
  {
  	struct net_device *dev = qdisc_dev(sch);
  	struct mqprio_sched *priv = qdisc_priv(sch);
-	unsigned char *b = skb_tail_pointer(skb);
+	struct nlattr *nla = (struct nlattr *)skb_tail_pointer(skb);
  	struct tc_mqprio_qopt opt = { 0 };
  	struct Qdisc *qdisc;
  	unsigned int i;
@@ -258,12 +400,17 @@ static int mqprio_dump(struct Qdisc *sch, struct sk_buff *skb)
  		opt.offset[i] = dev->tc_to_txq[i].offset;
  	}
 
-	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
+	if (nla_put(skb, TCA_OPTIONS, NLA_ALIGN(sizeof(opt)), &opt))
  		goto nla_put_failure;
 
-	return skb->len;
+	if (priv->flags & TC_MQPRIO_F_MIN_RATE ||
+	    priv->flags & TC_MQPRIO_F_MAX_RATE)
+		if (dump_rates(priv, &opt, skb) != 0)
+			goto nla_put_failure;
+
+	return nla_nest_end(skb, nla);
  nla_put_failure:
-	nlmsg_trim(skb, b);
+	nlmsg_trim(skb, nla);
  	return -1;
  }
 

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

#3 lore raw

PATCH v2 1/6 net: mqprio: Introduce new hardware offload mode in mqprio for offloading full TC configurations

Quoted 122 lines · Shannon Nelson

Two little thoughts...
sln

Quoted 118 lines · Amritha Nambiar

This patch introduces a new hardware offload mode in mqprio which makes full use of the mqprio options, the TCs, the queue configurations and the bandwidth rates for the TCs. This is achieved by setting the value 2 for the "hw" option. This new offload mode supports new attributes for traffic class such as minimum and maximum values for bandwidth rate limits.

Introduces a new datastructure 'tc_mqprio_qopt_offload' for offloading mqprio queue options and use this to be shared between the kernel and device driver. This contains a copy of the exisiting datastructure for mqprio queue options. This new datastructure can be extended when adding new attributes for traffic class such as bandwidth rate limits. The existing datastructure for mqprio queue options will be shared between the kernel and userspace.

This patch enables configuring additional attributes associated with a traffic class such as minimum and maximum bandwidth rates and can be offloaded to the hardware in the new offload mode. The min and max limits for bandwidth rates are provided by the user along with the the TCs and the queue configurations when creating the mqprio qdisc.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\
   queues 4 at 0 4 at 4 min_rate 10Mbit 20Mbit max_rate 55Mbit 60Mbit hw 2

To dump the bandwidth rates:

# tc qdisc show dev eth0

qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0

              queues:(0:3) (4:7)
              min_rate:10Mbit 20Mbit
              max_rate:55Mbit 60Mbit

v2 : Jiri's changes accepted upstream removes the struct tc_to_netdev. Clean up the full offload related changes added to mqprio_init() and mqprio_destroy() to rebase on these changes.

---
  include/linux/netdevice.h      |    1
  include/net/pkt_cls.h          |    7 ++
  include/uapi/linux/pkt_sched.h |   13 +++
  net/sched/sch_mqprio.c         |  167 ++++++++++++++++++++++++++++++++++++++--
  4 files changed, 178 insertions(+), 10 deletions(-)
include/linux/netdevice.h +1 -0
index 0f1c4cb..631f314 100644
@@ -777,6 +777,7 @@ enum tc_setup_type {
  	TC_SETUP_CLSFLOWER,
  	TC_SETUP_CLSMATCHALL,
  	TC_SETUP_CLSBPF,
+	TC_SETUP_MQPRIO_EXT,
  };
 
  /* These structures hold the attributes of xdp state that are being passed
include/net/pkt_cls.h +7 -0
index e80edd8..55bf148 100644
@@ -546,6 +546,13 @@ struct tc_cls_bpf_offload {
  	u32 gen_flags;
  };
 
+struct tc_mqprio_qopt_offload {
+	/* struct tc_mqprio_qopt must always be the first element */
+	struct tc_mqprio_qopt qopt;
+	u32 flags;
+	u64 min_rate[TC_QOPT_MAX_QUEUE];
+	u64 max_rate[TC_QOPT_MAX_QUEUE];
+};
 
  /* This structure holds cookie structure that is passed from user
   * to the kernel for actions and classifiers
include/uapi/linux/pkt_sched.h +13 -0
index 099bf55..cf2a146 100644
@@ -620,6 +620,7 @@ struct tc_drr_stats {
  enum {
  	TC_MQPRIO_HW_OFFLOAD_NONE,	/* no offload requested */
  	TC_MQPRIO_HW_OFFLOAD_TCS,	/* offload TCs, no queue counts */
+	TC_MQPRIO_HW_OFFLOAD,		/* fully supported offload */
  	__TC_MQPRIO_HW_OFFLOAD_MAX
  };
 
@@ -633,6 +634,18 @@ struct tc_mqprio_qopt {
  	__u16	offset[TC_QOPT_MAX_QUEUE];
  };
 
+#define TC_MQPRIO_F_MIN_RATE  0x1
+#define TC_MQPRIO_F_MAX_RATE  0x2
+
+enum {
+	TCA_MQPRIO_UNSPEC,
+	TCA_MQPRIO_MIN_RATE64,
+	TCA_MQPRIO_MAX_RATE64,
+	__TCA_MQPRIO_MAX,
+};
+
+#define TCA_MQPRIO_MAX (__TCA_MQPRIO_MAX - 1)
+
  /* SFB */
 
  enum {
net/sched/sch_mqprio.c +1 -0
index 2165a05..d913111 100644
@@ -18,10 +18,13 @@
  #include <net/netlink.h>
  #include <net/pkt_sched.h>
  #include <net/sch_generic.h>
+#include <net/pkt_cls.h>
 
Quoted 7 lines

net/sched/sch_mqprio.c line 27

Quoted 5 lines
net/sched/sch_mqprio.c +2 -0
  struct mqprio_sched {
  	struct Qdisc		**qdiscs;
  	int hw_offload;
+	u32 flags;
+	u64 min_rate[TC_QOPT_MAX_QUEUE], max_rate[TC_QOPT_MAX_QUEUE];

Put these on separate lines.

Will do in the next version.

Quoted 114 lines
Quoted 114 lines
net/sched/sch_mqprio.c +80 -2
  };
 
  static void mqprio_destroy(struct Qdisc *sch)
@@ -39,9 +42,20 @@ static void mqprio_destroy(struct Qdisc *sch)
  	}
 
  	if (priv->hw_offload && dev->netdev_ops->ndo_setup_tc) {
-		struct tc_mqprio_qopt mqprio = {};
+		struct tc_mqprio_qopt_offload mqprio = { { 0 } };
 
-		dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO, &mqprio);
+		switch (priv->hw_offload) {
+		case TC_MQPRIO_HW_OFFLOAD_TCS:
+			dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO,
+						      &mqprio.qopt);
+			break;
+		case TC_MQPRIO_HW_OFFLOAD:
+			dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO_EXT,
+						      &mqprio);
+			break;
+		default:
+			return;
+		}
  	} else {
  		netdev_set_num_tc(dev, 0);
  	}
@@ -97,6 +111,24 @@ static int mqprio_parse_opt(struct net_device *dev, struct tc_mqprio_qopt *qopt)
  	return 0;
  }
 
+static const struct nla_policy mqprio_policy[TCA_MQPRIO_MAX + 1] = {
+	[TCA_MQPRIO_MIN_RATE64] = { .type = NLA_NESTED },
+	[TCA_MQPRIO_MAX_RATE64] = { .type = NLA_NESTED },
+};
+
+static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
+		      const struct nla_policy *policy, int len)
+{
+	int nested_len = nla_len(nla) - NLA_ALIGN(len);
+
+	if (nested_len >= nla_attr_size(0))
+		return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
+				 nested_len, policy, NULL);
+
+	memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
+	return 0;
+}
+
  static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  {
  	struct net_device *dev = qdisc_dev(sch);
@@ -105,6 +137,10 @@ static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  	struct Qdisc *qdisc;
  	int i, err = -EOPNOTSUPP;
  	struct tc_mqprio_qopt *qopt = NULL;
+	struct nlattr *tb[TCA_MQPRIO_MAX + 1];
+	struct nlattr *attr;
+	int rem;
+	int len = nla_len(opt) - NLA_ALIGN(sizeof(*qopt));
 
  	BUILD_BUG_ON(TC_MAX_QUEUE != TC_QOPT_MAX_QUEUE);
  	BUILD_BUG_ON(TC_BITMASK != TC_QOPT_BITMASK);
@@ -122,6 +158,51 @@ static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  	if (mqprio_parse_opt(dev, qopt))
  		return -EINVAL;
 
+	if (len > 0) {
+		err = parse_attr(tb, TCA_MQPRIO_MAX, opt, mqprio_policy,
+				 sizeof(*qopt));
+		if (err < 0)
+			return err;
+
+		if (tb[TCA_MQPRIO_MIN_RATE64]) {
+			if (qopt->hw != TC_MQPRIO_HW_OFFLOAD)
+				return -EINVAL;
+
+			i = 0;
+			nla_for_each_nested(attr, tb[TCA_MQPRIO_MIN_RATE64],
+					    rem) {
+				if (nla_type(attr) != TCA_MQPRIO_MIN_RATE64)
+					return -EINVAL;
+
+				if (i >= qopt->num_tc)
+					return -EINVAL;
+
+				priv->min_rate[i] = *(u64 *)nla_data(attr);
+				i++;
+			}
+			priv->flags |= TC_MQPRIO_F_MIN_RATE;
+		}
+
+		if (tb[TCA_MQPRIO_MAX_RATE64]) {
+			if (qopt->hw != TC_MQPRIO_HW_OFFLOAD)
+				return -EINVAL;
+
+			i = 0;
+			nla_for_each_nested(attr, tb[TCA_MQPRIO_MAX_RATE64],
+					    rem) {
+				if (nla_type(attr) != TCA_MQPRIO_MAX_RATE64)
+					return -EINVAL;
+
+				if (i >= qopt->num_tc)
+					return -EINVAL;
+
+				priv->max_rate[i] = *(u64 *)nla_data(attr);
+				i++;
+			}
+			priv->flags |= TC_MQPRIO_F_MAX_RATE;
+		}
+	}
+
  	/* pre-allocate qdisc, attachment can't fail */
  	priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
Quoted 8 lines

net/sched/sch_mqprio.c static int mqprio_init(struct Qdisc *sch, struct nlattr *opt) · line 229

Quoted 5 lines
net/sched/sch_mqprio.c +0 -0
  			       GFP_KERNEL);
@@ -146,14 +227,35 @@ static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
  	 * supplied and verified mapping
  	 */
  	if (qopt->hw) {

In the changes above is a test for existence of ->ndo_setup_tc(), should there be a similar test here as well?

We wouldn't be here if ndo_setup_tc does not exist. It gets checked prior in mqprio_parse_opt() and returns if not found.

Quoted 117 lines
Quoted 117 lines
net/sched/sch_mqprio.c +74 -8
-		struct tc_mqprio_qopt mqprio = *qopt;
+		struct tc_mqprio_qopt_offload mqprio = {.qopt = *qopt};
 
-		err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_MQPRIO,
-						    &mqprio);
+		switch (qopt->hw) {
+		case TC_MQPRIO_HW_OFFLOAD_TCS:
+			err = dev->netdev_ops->ndo_setup_tc(dev,
+							    TC_SETUP_MQPRIO,
+							    &mqprio.qopt);
+			break;
+		case TC_MQPRIO_HW_OFFLOAD:
+			mqprio.flags = priv->flags;
+			if (priv->flags & TC_MQPRIO_F_MIN_RATE)
+				for (i = 0; i < mqprio.qopt.num_tc; i++)
+					mqprio.min_rate[i] = priv->min_rate[i];
+
+			if (priv->flags & TC_MQPRIO_F_MAX_RATE)
+				for (i = 0; i < mqprio.qopt.num_tc; i++)
+					mqprio.max_rate[i] = priv->max_rate[i];
+
+			err = dev->netdev_ops->ndo_setup_tc(dev,
+							    TC_SETUP_MQPRIO_EXT,
+							    &mqprio);
+			break;
+		default:
+			return -EINVAL;
+		}
  		if (err)
  			return err;
 
-		priv->hw_offload = mqprio.hw;
+		priv->hw_offload = mqprio.qopt.hw;
  	} else {
  		netdev_set_num_tc(dev, qopt->num_tc);
  		for (i = 0; i < qopt->num_tc; i++)
@@ -223,11 +325,51 @@ static int mqprio_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
  	return 0;
  }
 
+static int dump_rates(struct mqprio_sched *priv,
+		      struct tc_mqprio_qopt *opt, struct sk_buff *skb)
+{
+	struct nlattr *nest;
+	int i;
+
+	if (priv->flags & TC_MQPRIO_F_MIN_RATE) {
+		nest = nla_nest_start(skb, TCA_MQPRIO_MIN_RATE64);
+		if (!nest)
+			goto nla_put_failure;
+
+		for (i = 0; i < opt->num_tc; i++) {
+			if (nla_put(skb, TCA_MQPRIO_MIN_RATE64,
+				    sizeof(priv->min_rate[i]),
+				    &priv->min_rate[i]))
+				goto nla_put_failure;
+		}
+		nla_nest_end(skb, nest);
+	}
+
+	if (priv->flags & TC_MQPRIO_F_MAX_RATE) {
+		nest = nla_nest_start(skb, TCA_MQPRIO_MAX_RATE64);
+		if (!nest)
+			goto nla_put_failure;
+
+		for (i = 0; i < opt->num_tc; i++) {
+			if (nla_put(skb, TCA_MQPRIO_MAX_RATE64,
+				    sizeof(priv->max_rate[i]),
+				    &priv->max_rate[i]))
+				goto nla_put_failure;
+		}
+		nla_nest_end(skb, nest);
+	}
+	return 0;
+
+nla_put_failure:
+	nla_nest_cancel(skb, nest);
+	return -1;
+}
+
  static int mqprio_dump(struct Qdisc *sch, struct sk_buff *skb)
  {
  	struct net_device *dev = qdisc_dev(sch);
  	struct mqprio_sched *priv = qdisc_priv(sch);
-	unsigned char *b = skb_tail_pointer(skb);
+	struct nlattr *nla = (struct nlattr *)skb_tail_pointer(skb);
  	struct tc_mqprio_qopt opt = { 0 };
  	struct Qdisc *qdisc;
  	unsigned int i;
@@ -258,12 +400,17 @@ static int mqprio_dump(struct Qdisc *sch, struct sk_buff *skb)
  		opt.offset[i] = dev->tc_to_txq[i].offset;
  	}
 
-	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
+	if (nla_put(skb, TCA_OPTIONS, NLA_ALIGN(sizeof(opt)), &opt))
  		goto nla_put_failure;
 
-	return skb->len;
+	if (priv->flags & TC_MQPRIO_F_MIN_RATE ||
+	    priv->flags & TC_MQPRIO_F_MAX_RATE)
+		if (dump_rates(priv, &opt, skb) != 0)
+			goto nla_put_failure;
+
+	return nla_nest_end(skb, nla);
  nla_put_failure:
-	nlmsg_trim(skb, b);
+	nlmsg_trim(skb, nla);
  	return -1;
  }
 

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

#4 lore raw

PATCH v2 2/6 net: i40e: Add macro for PF reset bit

In reply to Amritha Nambiar

Introduce a macro for the bit setting the PF reset flag and update its usages. This makes it easier to use this flag in functions to be introduced in future without encountering checkpatch issues related to alignment and line over 80 characters.

---
 drivers/net/ethernet/intel/i40e/i40e.h             |    2 ++
 drivers/net/ethernet/intel/i40e/i40e_debugfs.c     |    3 +--
 drivers/net/ethernet/intel/i40e/i40e_main.c        |    9 ++++-----
 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c |    5 ++---
 4 files changed, 9 insertions(+), 10 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +2 -0
index 890508f..0e53838 100644
@@ -157,6 +157,8 @@ enum i40e_state_t {
 	__I40E_STATE_SIZE__,
 };
 
+#define I40E_PF_RESET_FLAG	BIT_ULL(__I40E_PF_RESET_REQUESTED)
+
 /* VSI state flags */
 enum i40e_vsi_state_t {
 	__I40E_VSI_DOWN,
drivers/net/ethernet/intel/i40e/i40e_debugfs.c +1 -2
index 8f326f8..b46117e 100644
@@ -798,8 +798,7 @@ static ssize_t i40e_dbg_command_write(struct file *filp,
 		 */
 		if (!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) {
 			pf->flags |= I40E_FLAG_VEB_MODE_ENABLED;
-			i40e_do_reset_safe(pf,
-					   BIT_ULL(__I40E_PF_RESET_REQUESTED));
+			i40e_do_reset_safe(pf, I40E_PF_RESET_FLAG);
 		}
 
 		vsi = i40e_vsi_setup(pf, I40E_VSI_VMDQ2, vsi_seid, 0);
drivers/net/ethernet/intel/i40e/i40e_main.c +4 -5
index d2ca98f..9392c5a 100644
@@ -5753,7 +5753,7 @@ int i40e_vsi_open(struct i40e_vsi *vsi)
 err_setup_tx:
 	i40e_vsi_free_tx_resources(vsi);
 	if (vsi == pf->vsi[pf->lan_vsi])
-		i40e_do_reset(pf, BIT_ULL(__I40E_PF_RESET_REQUESTED), true);
+		i40e_do_reset(pf, I40E_PF_RESET_FLAG, true);
 
 	return err;
 }
@@ -5881,7 +5881,7 @@ void i40e_do_reset(struct i40e_pf *pf, u32 reset_flags, bool lock_acquired)
 		wr32(&pf->hw, I40E_GLGEN_RTRIG, val);
 		i40e_flush(&pf->hw);
 
-	} else if (reset_flags & BIT_ULL(__I40E_PF_RESET_REQUESTED)) {
+	} else if (reset_flags & I40E_PF_RESET_FLAG) {
 
 		/* Request a PF Reset
 		 *
@@ -9209,7 +9209,7 @@ static int i40e_set_features(struct net_device *netdev,
 	need_reset = i40e_set_ntuple(pf, features);
 
 	if (need_reset)
-		i40e_do_reset(pf, BIT_ULL(__I40E_PF_RESET_REQUESTED), true);
+		i40e_do_reset(pf, I40E_PF_RESET_FLAG, true);
 
 	return 0;
 }
@@ -9461,8 +9461,7 @@ static int i40e_ndo_bridge_setlink(struct net_device *dev,
 				pf->flags |= I40E_FLAG_VEB_MODE_ENABLED;
 			else
 				pf->flags &= ~I40E_FLAG_VEB_MODE_ENABLED;
-			i40e_do_reset(pf, BIT_ULL(__I40E_PF_RESET_REQUESTED),
-				      true);
+			i40e_do_reset(pf, I40E_PF_RESET_FLAG, true);
 			break;
 		}
 	}
drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +2 -3
index e156096..92fe06d 100644
@@ -1400,8 +1400,7 @@ int i40e_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
 	if (num_vfs) {
 		if (!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) {
 			pf->flags |= I40E_FLAG_VEB_MODE_ENABLED;
-			i40e_do_reset_safe(pf,
-					   BIT_ULL(__I40E_PF_RESET_REQUESTED));
+			i40e_do_reset_safe(pf, I40E_PF_RESET_FLAG);
 		}
 		return i40e_pci_sriov_enable(pdev, num_vfs);
 	}
@@ -1409,7 +1408,7 @@ int i40e_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
 	if (!pci_vfs_assigned(pf->pdev)) {
 		i40e_free_vfs(pf);
 		pf->flags &= ~I40E_FLAG_VEB_MODE_ENABLED;
-		i40e_do_reset_safe(pf, BIT_ULL(__I40E_PF_RESET_REQUESTED));
+		i40e_do_reset_safe(pf, I40E_PF_RESET_FLAG);
 	} else {
 		dev_warn(&pdev->dev, "Unable to free VFs because some are assigned to VMs.\n");
 		return -EINVAL;
#5 lore raw

PATCH v2 3/6 net: i40e: Add infrastructure for queue channel support with the TCs and queue configurations offloaded via mqprio scheduler

In reply to Amritha Nambiar

This patch sets up the infrastructure for offloading TCs and queue configurations to the hardware by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0). TC0 for the main VSI is also reconfigured as per user provided queue parameters. Queue counts that are not power-of-2 are handled by reconfiguring RSS by reprogramming LUTs using the queue count value. This patch also handles configuring the TX rings for the channels, setting up the RX queue map for channel.

Also, the channels so created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.

---
 drivers/net/ethernet/intel/i40e/i40e.h      |   33 +
 drivers/net/ethernet/intel/i40e/i40e_main.c |  747 +++++++++++++++++++++++++++
 drivers/net/ethernet/intel/i40e/i40e_txrx.h |    2 
 3 files changed, 773 insertions(+), 9 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +33 -0
index 0e53838..e259a67 100644
@@ -87,6 +87,7 @@
 #define I40E_AQ_LEN			256
 #define I40E_AQ_WORK_LIMIT		66 /* max number of VFs + a little */
 #define I40E_MAX_USER_PRIORITY		8
+#define I40E_MAX_QUEUES_PER_CH		64
 #define I40E_DEFAULT_TRAFFIC_CLASS	BIT(0)
 #define I40E_DEFAULT_MSG_ENABLE		4
 #define I40E_QUEUE_WAIT_RETRY_LIMIT	10
@@ -340,6 +341,23 @@ struct i40e_flex_pit {
 	u8 pit_index;
 };
 
+struct i40e_channel {
+	struct list_head list;
+	bool initialized;
+	u8 type;
+	u16 vsi_number; /* Assigned VSI number from AQ 'Add VSI' response */
+	u16 stat_counter_idx;
+	u16 base_queue;
+	u16 num_queue_pairs; /* Requested by user */
+	u16 seid;
+
+	u8 enabled_tc;
+	struct i40e_aqc_vsi_properties_data info;
+
+	/* track this channel belongs to which VSI */
+	struct i40e_vsi *parent_vsi;
+};
+
 /* struct that defines the Ethernet device */
 struct i40e_pf {
 	struct pci_dev *pdev;
@@ -455,6 +473,7 @@ struct i40e_pf {
 #define I40E_FLAG_CLIENT_L2_CHANGE		BIT_ULL(25)
 #define I40E_FLAG_CLIENT_RESET			BIT_ULL(26)
 #define I40E_FLAG_LINK_DOWN_ON_CLOSE_ENABLED	BIT_ULL(27)
+#define I40E_FLAG_TC_MQPRIO			BIT_ULL(28)
 
 	struct i40e_client_instance *cinst;
 	bool stat_offsets_loaded;
@@ -535,6 +554,8 @@ struct i40e_pf {
 	u32 ioremap_len;
 	u32 fd_inv;
 	u16 phy_led_val;
+
+	u16 override_q_count;
 };
 
 /**
@@ -699,6 +720,16 @@ struct i40e_vsi {
 	bool current_isup;	/* Sync 'link up' logging */
 	enum i40e_aq_link_speed current_speed;	/* Sync link speed logging */
 
+	/* channel specific fields */
+	u16 cnt_q_avail; /* num of queues available for channel usage */
+	u16 orig_rss_size;
+	u16 current_rss_size;
+
+	/* keeps track of next_base_queue to be used for channel setup */
+	atomic_t next_base_queue;
+
+	struct list_head ch_list;
+
 	void *priv;	/* client driver data reference. */
 
 	/* VSI specific handlers */
@@ -1006,4 +1037,6 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi)
 {
 	return !!vsi->xdp_prog;
 }
+
+int i40e_create_queue_channel(struct i40e_vsi *vsi, struct i40e_channel *ch);
 #endif /* _I40E_H_ */
drivers/net/ethernet/intel/i40e/i40e_main.c +738 -9

798 lines not on this page · raw message

drivers/net/ethernet/intel/i40e/i40e_txrx.h +2 -0
index 2f848bc..a767f87 100644
@@ -423,6 +423,8 @@ struct i40e_ring {
 					 * i40e_clean_rx_ring_irq() is called
 					 * for this ring.
 					 */
+
+	struct i40e_channel *ch;
 } ____cacheline_internodealigned_in_smp;
 
 static inline bool ring_uses_build_skb(struct i40e_ring *ring)
#6 lore raw

PATCH v2 3/6 net: i40e: Add infrastructure for queue channel support with the TCs and queue configurations offloaded via mqprio scheduler

Quoted 10 lines · Amritha Nambiar

This patch sets up the infrastructure for offloading TCs and queue configurations to the hardware by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0). TC0 for the main VSI is also reconfigured as per user provided queue parameters. Queue counts that are not power-of-2 are handled by reconfiguring RSS by reprogramming LUTs using the queue count value. This patch also handles configuring the TX rings for the channels, setting up the RX queue map for channel.

Quoted 3 lines

Also, the channels so created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.

To reference an old pet-peave of mine, these channels might be how to support macvlan offloads - you might look into how these can support ndo_dfwd_add_station and ndo_dfwd_del_station.

Quoted 285 lines
---
  drivers/net/ethernet/intel/i40e/i40e.h      |   33 +
  drivers/net/ethernet/intel/i40e/i40e_main.c |  747 +++++++++++++++++++++++++++
  drivers/net/ethernet/intel/i40e/i40e_txrx.h |    2
  3 files changed, 773 insertions(+), 9 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +33 -0
index 0e53838..e259a67 100644
@@ -87,6 +87,7 @@
  #define I40E_AQ_LEN			256
  #define I40E_AQ_WORK_LIMIT		66 /* max number of VFs + a little */
  #define I40E_MAX_USER_PRIORITY		8
+#define I40E_MAX_QUEUES_PER_CH		64
  #define I40E_DEFAULT_TRAFFIC_CLASS	BIT(0)
  #define I40E_DEFAULT_MSG_ENABLE		4
  #define I40E_QUEUE_WAIT_RETRY_LIMIT	10
@@ -340,6 +341,23 @@ struct i40e_flex_pit {
  	u8 pit_index;
  };
 
+struct i40e_channel {
+	struct list_head list;
+	bool initialized;
+	u8 type;
+	u16 vsi_number; /* Assigned VSI number from AQ 'Add VSI' response */
+	u16 stat_counter_idx;
+	u16 base_queue;
+	u16 num_queue_pairs; /* Requested by user */
+	u16 seid;
+
+	u8 enabled_tc;
+	struct i40e_aqc_vsi_properties_data info;
+
+	/* track this channel belongs to which VSI */
+	struct i40e_vsi *parent_vsi;
+};
+
  /* struct that defines the Ethernet device */
  struct i40e_pf {
  	struct pci_dev *pdev;
@@ -455,6 +473,7 @@ struct i40e_pf {
  #define I40E_FLAG_CLIENT_L2_CHANGE		BIT_ULL(25)
  #define I40E_FLAG_CLIENT_RESET			BIT_ULL(26)
  #define I40E_FLAG_LINK_DOWN_ON_CLOSE_ENABLED	BIT_ULL(27)
+#define I40E_FLAG_TC_MQPRIO			BIT_ULL(28)
 
  	struct i40e_client_instance *cinst;
  	bool stat_offsets_loaded;
@@ -535,6 +554,8 @@ struct i40e_pf {
  	u32 ioremap_len;
  	u32 fd_inv;
  	u16 phy_led_val;
+
+	u16 override_q_count;
  };
 
  /**
@@ -699,6 +720,16 @@ struct i40e_vsi {
  	bool current_isup;	/* Sync 'link up' logging */
  	enum i40e_aq_link_speed current_speed;	/* Sync link speed logging */
 
+	/* channel specific fields */
+	u16 cnt_q_avail; /* num of queues available for channel usage */
+	u16 orig_rss_size;
+	u16 current_rss_size;
+
+	/* keeps track of next_base_queue to be used for channel setup */
+	atomic_t next_base_queue;
+
+	struct list_head ch_list;
+
  	void *priv;	/* client driver data reference. */
 
  	/* VSI specific handlers */
@@ -1006,4 +1037,6 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi)
  {
  	return !!vsi->xdp_prog;
  }
+
+int i40e_create_queue_channel(struct i40e_vsi *vsi, struct i40e_channel *ch);
  #endif /* _I40E_H_ */
drivers/net/ethernet/intel/i40e/i40e_main.c +162 -7
index 9392c5a..64769cb 100644
@@ -2887,7 +2887,7 @@ static void i40e_config_xps_tx_ring(struct i40e_ring *ring)
  {
  	struct i40e_vsi *vsi = ring->vsi;
 
-	if (!ring->q_vector || !ring->netdev)
+	if (!ring->q_vector || !ring->netdev || ring->ch)
  		return;
 
  	if ((vsi->tc_config.numtc <= 1) &&
@@ -2954,7 +2954,14 @@ static int i40e_configure_tx_ring(struct i40e_ring *ring)
  	 * initialization. This has to be done regardless of
  	 * DCB as by default everything is mapped to TC0.
  	 */
-	tx_ctx.rdylist = le16_to_cpu(vsi->info.qs_handle[ring->dcb_tc]);
+
+	if (ring->ch)
+		tx_ctx.rdylist =
+			le16_to_cpu(ring->ch->info.qs_handle[ring->dcb_tc]);
+
+	else
+		tx_ctx.rdylist = le16_to_cpu(vsi->info.qs_handle[ring->dcb_tc]);
+
  	tx_ctx.rdylist_act = 0;
 
  	/* clear the context in the HMC */
@@ -2976,12 +2983,23 @@ static int i40e_configure_tx_ring(struct i40e_ring *ring)
  	}
 
  	/* Now associate this queue with this PCI function */
-	if (vsi->type == I40E_VSI_VMDQ2) {
-		qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
-		qtx_ctl |= ((vsi->id) << I40E_QTX_CTL_VFVM_INDX_SHIFT) &
-			   I40E_QTX_CTL_VFVM_INDX_MASK;
+	if (ring->ch) {
+		if (ring->ch->type == I40E_VSI_VMDQ2)
+			qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
+		else
+			return -EINVAL;
+
+		qtx_ctl |= (ring->ch->vsi_number <<
+			    I40E_QTX_CTL_VFVM_INDX_SHIFT) &
+			    I40E_QTX_CTL_VFVM_INDX_MASK;
  	} else {
-		qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
+		if (vsi->type == I40E_VSI_VMDQ2) {
+			qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
+			qtx_ctl |= ((vsi->id) << I40E_QTX_CTL_VFVM_INDX_SHIFT) &
+				    I40E_QTX_CTL_VFVM_INDX_MASK;
+		} else {
+			qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
+		}
  	}
 
  	qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
@@ -5168,6 +5186,672 @@ static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  }
 
  /**
+ * i40e_remove_queue_channels - Remove queue channels for the TCs
+ * @vsi: VSI to be configured
+ *
+ * Remove queue channels for the TCs
+ **/
+static void i40e_remove_queue_channels(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+	int ret, i;
+
+	/* Reset rss size that was stored when reconfiguring rss for
+	 * channel VSIs with non-power-of-2 queue count.
+	 */
+	vsi->current_rss_size = 0;
+
+	/* perform cleanup for channels if they exist */
+	if (list_empty(&vsi->ch_list))
+		return;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		struct i40e_vsi *p_vsi;
+
+		list_del(&ch->list);
+		p_vsi = ch->parent_vsi;
+		if (!p_vsi || !ch->initialized) {
+			kfree(ch);
+			continue;
+		}
+		/* Reset queue contexts */
+		for (i = 0; i < ch->num_queue_pairs; i++) {
+			struct i40e_ring *tx_ring, *rx_ring;
+			u16 pf_q;
+
+			pf_q = ch->base_queue + i;
+			tx_ring = vsi->tx_rings[pf_q];
+			tx_ring->ch = NULL;
+
+			rx_ring = vsi->rx_rings[pf_q];
+			rx_ring->ch = NULL;
+		}
+
+		/* delete VSI from FW */
+		ret = i40e_aq_delete_element(&vsi->back->hw, ch->seid,
+					     NULL);
+		if (ret)
+			dev_err(&vsi->back->pdev->dev,
+				"unable to remove channel (%d) for parent VSI(%d)\n",
+				ch->seid, p_vsi->seid);
+		kfree(ch);
+	}
+	INIT_LIST_HEAD(&vsi->ch_list);
+}
+
+/**
+ * i40e_is_any_channel - channel exist or not
+ * @vsi: ptr to VSI to which channels are associated with
+ *
+ * Returns true or false if channel(s) exist for associated VSI or not
+ **/
+static bool i40e_is_any_channel(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		if (ch->initialized)
+			return true;
+	}
+
+	return false;
+}
+
+/**
+ * i40e_get_max_queues_for_channel
+ * @vsi: ptr to VSI to which channels are associated with
+ *
+ * Helper function which returns max value among the queue counts set on the
+ * channels/TCs created.
+ **/
+static int i40e_get_max_queues_for_channel(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+	int max = 0;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		if (!ch->initialized)
+			continue;
+		if (ch->num_queue_pairs > max)
+			max = ch->num_queue_pairs;
+	}
+
+	return max;
+}
+
+/**
+ * i40e_validate_num_queues - validate num_queues w.r.t channel
+ * @pf: ptr to PF device
+ * @num_queues: number of queues
+ * @vsi: the parent VSI
+ * @reconfig_rss: indicates should the RSS be reconfigured or not
+ *
+ * This function validates number of queues in the context of new channel
+ * which is being established and determines if RSS should be reconfigured
+ * or not for parent VSI.
+ **/
+static int i40e_validate_num_queues(struct i40e_pf *pf, int num_queues,
+				    struct i40e_vsi *vsi, bool *reconfig_rss)
+{
+	int max_ch_queues;
+
+	if (!reconfig_rss)
+		return -EINVAL;
+
+	*reconfig_rss = false;
+
+	if (num_queues > I40E_MAX_QUEUES_PER_CH) {
+		dev_err(&pf->pdev->dev,
+			"Failed to create VMDq VSI. User requested num_queues (%d) > I40E_MAX_QUEUES_PER_VSI (%u)\n",
+			num_queues, I40E_MAX_QUEUES_PER_CH);
+		return -EINVAL;
+	}
+
+	if (vsi->current_rss_size) {
+		if (num_queues > vsi->current_rss_size) {
+			dev_dbg(&pf->pdev->dev,
+				"Error: num_queues (%d) > vsi's current_size(%d)\n",
+				num_queues, vsi->current_rss_size);
+			return -EINVAL;
+		} else if ((num_queues < vsi->current_rss_size) &&
+			   (!is_power_of_2(num_queues))) {
+			dev_dbg(&pf->pdev->dev,
+				"Error: num_queues (%d) < vsi's current_size(%d), but not power of 2\n",
+				num_queues, vsi->current_rss_size);
+			return -EINVAL;
+		}
+	}
+
+	if (!is_power_of_2(num_queues)) {

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5335

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+		/* Find the max num_queues configures for channel if channel
+		 * exist.
+		 * if channel exist, then enforce 'num_queues' to be more than
+		 * max ever num_queues configured for channel.
+		 */
+		max_ch_queues = i40e_get_max_queues_for_channel(vsi);
+		if (num_queues < max_ch_queues) {
+			dev_dbg(&pf->pdev->dev,
+				"Error: num_queues (%d) > main_vsi's current_size(%d)\n",
+				num_queues, vsi->current_rss_size);

This doesn't look like it matches the if expression - should the expression be '>' rather than '<'?

Quoted 75 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +75 -0
+			return -EINVAL;
+		}
+		*reconfig_rss = true;
+	}
+
+	return 0;
+}
+
+/**
+ * i40e_vsi_reconfig_rss - reconfig RSS based on specified rss_size
+ * @vsi: the VSI being setup
+ * @rss_size: size of RSS, accordingly LUT gets reprogrammed
+ *
+ * This function reconfigures RSS by reprogramming LUTs using 'rss_size'
+ **/
+static int i40e_vsi_reconfig_rss(struct i40e_vsi *vsi, u16 rss_size)
+{
+	struct i40e_pf *pf = vsi->back;
+	u8 seed[I40E_HKEY_ARRAY_SIZE];
+	struct i40e_hw *hw = &pf->hw;
+	int local_rss_size;
+	u8 *lut;
+	int ret;
+
+	if (!vsi->rss_size)
+		return -EINVAL;
+
+	if (rss_size > vsi->rss_size)
+		return -EINVAL;
+
+	local_rss_size = min_t(int, vsi->rss_size, rss_size);
+	lut = kzalloc(vsi->rss_table_size, GFP_KERNEL);
+	if (!lut)
+		return -ENOMEM;
+
+	/* Ignoring user configured lut if there is one */
+	i40e_fill_rss_lut(pf, lut, vsi->rss_table_size, local_rss_size);
+
+	/* Use user configured hash key if there is one, otherwise
+	 * use default.
+	 */
+	if (vsi->rss_hkey_user)
+		memcpy(seed, vsi->rss_hkey_user, I40E_HKEY_ARRAY_SIZE);
+	else
+		netdev_rss_key_fill((void *)seed, I40E_HKEY_ARRAY_SIZE);
+
+	ret = i40e_config_rss(vsi, seed, lut, vsi->rss_table_size);
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "Cannot set RSS lut, err %s aq_err %s\n",
+			 i40e_stat_str(hw, ret),
+			 i40e_aq_str(hw, hw->aq.asq_last_status));
+		kfree(lut);
+		return ret;
+	}
+	kfree(lut);
+
+	/* Do the update w.r.t. storing rss_size */
+	if (!vsi->orig_rss_size)
+		vsi->orig_rss_size = vsi->rss_size;
+	vsi->current_rss_size = local_rss_size;
+
+	return ret;
+}
+
+/**
+ * i40e_channel_setup_queue_map - Setup a channel queue map based on enabled_tc
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ctxt: VSI context structure
+ * @enabled_tc: Enabled TCs bitmap
+ * @ch: ptr to channel structure
+ *
+ * Setup queue map based on enabled_tc for specific channel
+ **/

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5420

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+static void i40e_channel_setup_queue_map(struct i40e_pf *pf,
+					 struct i40e_vsi_context *ctxt,
+					 u8 enabled_tc, struct i40e_channel *ch)
+{
+	u16 qcount, num_tc_qps, qmap;
+	int pow, num_qps;
+	u16 sections = 0;
+	/* At least TC0 is enabled in case of non-DCB case, non-MQPRIO */
+	u16 numtc = 1;
+	u8 offset = 0;

The enabled_tc parameter and the numtc variable are useless here, I assume they'll be useful in a later patch?

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5431

Quoted 11 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +11 -0
+
+	sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID;
+	sections |= I40E_AQ_VSI_PROP_SCHED_VALID;
+
+	if (pf->flags & I40E_FLAG_MSIX_ENABLED) {
+		qcount = min_t(int, ch->num_queue_pairs,
+			       pf->num_lan_msix);
+		ch->num_queue_pairs = qcount;
+	} else {
+		qcount = 1;
+	}

If no MSIX, we probably shouldn't even be here.

Quoted 206 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +206 -0
+
+	/* find num of qps per traffic class */
+	num_tc_qps = qcount / numtc;
+	num_tc_qps = min_t(int, num_tc_qps, i40e_pf_get_max_q_per_tc(pf));
+	num_qps = qcount;
+
+	/* find the next higher power-of-2 of num queue pairs */
+	pow = ilog2(num_qps);
+	if (!is_power_of_2(num_qps))
+		pow++;
+
+	qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) |
+		(pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT);
+
+	/* Setup queue TC[0].qmap for given VSI context */
+	ctxt->info.tc_mapping[0] = cpu_to_le16(qmap);
+
+	ctxt->info.up_enable_bits = enabled_tc;
+	ctxt->info.mapping_flags |= cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG);
+	ctxt->info.queue_mapping[0] = cpu_to_le16(ch->base_queue);
+	ctxt->info.valid_sections |= cpu_to_le16(sections);
+}
+
+/**
+ * i40e_add_channel - add a channel by adding VSI
+ * @pf: ptr to PF device
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @ch: ptr to channel structure
+ *
+ * Add a channel (VSI) using add_vsi and queue_map
+ **/
+static int i40e_add_channel(struct i40e_pf *pf, u16 uplink_seid,
+			    struct i40e_channel *ch)
+{
+	struct i40e_hw *hw = &pf->hw;
+	struct i40e_vsi_context ctxt;
+	u8 enabled_tc = 0x1; /* TC0 enabled */
+	int ret;
+
+	if (ch->type != I40E_VSI_VMDQ2) {
+		dev_info(&pf->pdev->dev,
+			 "add new vsi failed, ch->type %d\n", ch->type);
+		return -EINVAL;
+	}
+
+	memset(&ctxt, 0, sizeof(ctxt));
+	ctxt.pf_num = hw->pf_id;
+	ctxt.vf_num = 0;
+	ctxt.uplink_seid = uplink_seid;
+	ctxt.connection_type = I40E_AQ_VSI_CONN_TYPE_NORMAL;
+	if (ch->type == I40E_VSI_VMDQ2)
+		ctxt.flags = I40E_AQ_VSI_TYPE_VMDQ2;
+
+	if (pf->flags & I40E_FLAG_VEB_MODE_ENABLED) {
+		ctxt.info.valid_sections |=
+		     cpu_to_le16(I40E_AQ_VSI_PROP_SWITCH_VALID);
+		ctxt.info.switch_id =
+		   cpu_to_le16(I40E_AQ_VSI_SW_ID_FLAG_ALLOW_LB);
+	}
+
+	/* Set queue map for a given VSI context */
+	i40e_channel_setup_queue_map(pf, &ctxt, enabled_tc, ch);
+
+	/* Now time to create VSI */
+	ret = i40e_aq_add_vsi(hw, &ctxt, NULL);
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "add new vsi failed, err %s aq_err %s\n",
+			 i40e_stat_str(&pf->hw, ret),
+			 i40e_aq_str(&pf->hw,
+				     pf->hw.aq.asq_last_status));
+		return -ENOENT;
+	}
+
+	/* Success, update channel */
+	ch->enabled_tc = enabled_tc;
+	ch->seid = ctxt.seid;
+	ch->vsi_number = ctxt.vsi_number;
+	ch->stat_counter_idx = cpu_to_le16(ctxt.info.stat_counter_idx);
+
+	/* copy just the sections touched not the entire info
+	 * since not all sections are valid as returned by
+	 * update vsi params
+	 */
+	ch->info.mapping_flags = ctxt.info.mapping_flags;
+	memcpy(&ch->info.queue_mapping,
+	       &ctxt.info.queue_mapping, sizeof(ctxt.info.queue_mapping));
+	memcpy(&ch->info.tc_mapping, ctxt.info.tc_mapping,
+	       sizeof(ctxt.info.tc_mapping));
+
+	return 0;
+}
+
+static int i40e_channel_config_bw(struct i40e_vsi *vsi, struct i40e_channel *ch,
+				  u8 *bw_share)
+{
+	struct i40e_aqc_configure_vsi_tc_bw_data bw_data;
+	i40e_status ret;
+	int i;
+
+	bw_data.tc_valid_bits = ch->enabled_tc;
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
+		bw_data.tc_bw_credits[i] = bw_share[i];
+
+	ret = i40e_aq_config_vsi_tc_bw(&vsi->back->hw, ch->seid,
+				       &bw_data, NULL);
+	if (ret) {
+		dev_info(&vsi->back->pdev->dev,
+			 "Config VSI BW allocation per TC failed, aq_err: %d for new_vsi->seid %u\n",
+			 vsi->back->hw.aq.asq_last_status, ch->seid);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
+		ch->info.qs_handle[i] = bw_data.qs_handles[i];
+
+	return 0;
+}
+
+/**
+ * i40e_channel_config_tx_ring - config TX ring associated with new channel
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ch: ptr to channel structure
+ *
+ * Configure TX rings associated with channel (VSI) since queues are being
+ * from parent VSI.
+ **/
+static int i40e_channel_config_tx_ring(struct i40e_pf *pf,
+				       struct i40e_vsi *vsi,
+				       struct i40e_channel *ch)
+{
+	i40e_status ret;
+	int i;
+	u8 bw_share[I40E_MAX_TRAFFIC_CLASS] = {0};
+
+	/* Enable ETS TCs with equal BW Share for now across all VSIs */
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+		if (ch->enabled_tc & BIT(i))
+			bw_share[i] = 1;
+	}
+
+	/* configure BW for new VSI */
+	ret = i40e_channel_config_bw(vsi, ch, bw_share);
+	if (ret) {
+		dev_info(&vsi->back->pdev->dev,
+			 "Failed configuring TC map %d for channel (seid %u)\n",
+			 ch->enabled_tc, ch->seid);
+		return ret;
+	}
+
+	for (i = 0; i < ch->num_queue_pairs; i++) {
+		struct i40e_ring *tx_ring, *rx_ring;
+		u16 pf_q;
+
+		pf_q = ch->base_queue + i;
+
+		/* Get to TX ring ptr of main VSI, for re-setup TX queue
+		 * context
+		 */
+		tx_ring = vsi->tx_rings[pf_q];
+		tx_ring->ch = ch;
+
+		/* Get the RX ring ptr */
+		rx_ring = vsi->rx_rings[pf_q];
+		rx_ring->ch = ch;
+	}
+
+	return 0;
+}
+
+/**
+ * i40e_setup_hw_channel - setup new channel
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ch: ptr to channel structure
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @type: type of channel to be created (VMDq2/VF)
+ *
+ * Setup new channel (VSI) based on specified type (VMDq2/VF)
+ * and configures TX rings accordingly
+ **/
+static inline int i40e_setup_hw_channel(struct i40e_pf *pf,
+					struct i40e_vsi *vsi,
+					struct i40e_channel *ch,
+					u16 uplink_seid, u8 type)
+{
+	struct i40e_q_vector *q_vector;
+	int base_queue = 0;
+	int ret, i;
+
+	ch->initialized = false;
+	ch->base_queue = atomic_read(&vsi->next_base_queue);
+	ch->type = type;
+
+	/* Proceed with creation of channel (VMDq2) VSI */
+	ret = i40e_add_channel(pf, uplink_seid, ch);
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "failed to add_channel using uplink_seid %u\n",
+			 uplink_seid);
+		return ret;
+	}
+
+	/* Mark the successful creation of channel */
+	ch->initialized = true;

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5647

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+
+	/* Mark q_vectors indicating that they are part of newly created
+	 * channel (VSI)
+	 */
+	base_queue = ch->base_queue;
+	for (i = 0; i < ch->num_queue_pairs; i++) {
+		q_vector = vsi->tx_rings[base_queue + i]->q_vector;
+
+		if (!q_vector)
+			continue;

This seems like a fairly useless for-loop. Perhaps a later patch adds something here?

Quoted 4 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +4 -0
+	}
+
+	/* Reconfigure TX queues using QTX_CTL register */
+	ret = i40e_channel_config_tx_ring(pf, vsi, ch);

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5661

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "failed to configure TX rings for channel %u\n",
+			 ch->seid);
+		return ret;
+	}
+
+	/* update 'next_base_queue' */
+	atomic_set(&vsi->next_base_queue,
+		   atomic_read(&vsi->next_base_queue) + ch->num_queue_pairs);

Is there any possibility that this can get changed by a separate thread in between getting the base_queue and here? If not, why bother with atomics? If so, shouldn't there be a lock between here and there?

Quoted 14 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +14 -0
+
+	dev_dbg(&pf->pdev->dev,
+		"Added channel: vsi_seid %u, vsi_number %u, stat_counter_idx %u, num_queue_pairs %u, pf->next_base_queue %d\n",
+		ch->seid, ch->vsi_number, ch->stat_counter_idx,
+		ch->num_queue_pairs,
+		atomic_read(&vsi->next_base_queue));
+
+	return ret;
+}
+
+/**
+ * i40e_setup_channel - setup new channel using uplink element
+ * @pf: ptr to PF device
+ * @type: type of channel to be created (VMDq2/VF)

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5685

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @ch: ptr to channel structure
+ *
+ * Setup new channel (VSI) based on specified type (VMDq2/VF)
+ * and uplink switching element (uplink_seid)
+ **/
+static bool i40e_setup_channel(struct i40e_pf *pf, struct i40e_vsi *vsi,
+			       struct i40e_channel *ch)
+{
+	u16 seid = vsi->seid;

Why set this when it gets set differently in a couple of lines below?

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5692

Quoted 7 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +7 -0
+	u8 vsi_type;
+	int ret;
+
+	if (vsi->type == I40E_VSI_MAIN) {
+		vsi_type = I40E_VSI_VMDQ2;
+	} else {
+		dev_err(&pf->pdev->dev, "unsupported vsi type(%d) of parent vsi\n",

Might be better worded as "unsupported parent vsi type (%d)"

Quoted 74 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +74 -0
+			vsi->type);
+		return false;
+	} > +
+	/* underlying switching element */
+	seid = pf->vsi[pf->lan_vsi]->uplink_seid;
+
+	/* create channel (VSI), configure TX rings */
+	ret = i40e_setup_hw_channel(pf, vsi, ch, seid, vsi_type);
+	if (ret) {
+		dev_err(&pf->pdev->dev, "failed to setup hw_channel\n");
+		return false;
+	}
+
+	return ch->initialized ? true : false;
+}
+
+/**
+ * i40e_create_queue_channel - function to create channel
+ * @vsi: VSI to be configured
+ * @ch: ptr to channel (it contains channel specific params)
+ *
+ * This function creates channel (VSI) using num_queues specified by user,
+ * reconfigs RSS if needed.
+ **/
+int i40e_create_queue_channel(struct i40e_vsi *vsi,
+			      struct i40e_channel *ch)
+{
+	struct i40e_pf *pf = vsi->back;
+	bool reconfig_rss;
+	int err;
+
+	if (!ch)
+		return -EINVAL;
+
+	if (!ch->num_queue_pairs) {
+		dev_err(&pf->pdev->dev, "Invalid num_queues requested: %d\n",
+			ch->num_queue_pairs);
+		return -EINVAL;
+	}
+
+	/* validate user requested num_queues for channel */
+	err = i40e_validate_num_queues(pf, ch->num_queue_pairs, vsi,
+				       &reconfig_rss);
+	if (err) {
+		dev_info(&pf->pdev->dev, "Failed to validate num_queues (%d)\n",
+			 ch->num_queue_pairs);
+		return -EINVAL;
+	}
+
+	/* By default we are in VEPA mode, if this is the first VF/VMDq
+	 * VSI to be added switch to VEB mode.
+	 */
+	if ((!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) ||
+	    (!i40e_is_any_channel(vsi))) {
+		if (!is_power_of_2(vsi->tc_config.tc_info[0].qcount)) {
+			dev_dbg(&pf->pdev->dev,
+				"Failed to create channel. Override queues (%u) not power of 2\n",
+				vsi->tc_config.tc_info[0].qcount);
+			return -EINVAL;
+		}
+
+		if (!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) {
+			pf->flags |= I40E_FLAG_VEB_MODE_ENABLED;
+
+			if (vsi->type == I40E_VSI_MAIN) {
+				if (pf->flags & I40E_FLAG_TC_MQPRIO)
+					i40e_do_reset(pf, I40E_PF_RESET_FLAG,
+						      true);
+				else
+					i40e_do_reset_safe(pf,
+							   I40E_PF_RESET_FLAG);
+			}
+		}
+		/* now onwards for main VSI, number of queues will be value

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5777

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+		 * of TC0's queue count
+		 */
+	}
+
+	/* By this time, vsi->cnt_q_avail shall be set to non-zero and
+	 * it should be more than num_queues
+	 */
+	if (!vsi->cnt_q_avail || (vsi->cnt_q_avail < ch->num_queue_pairs)) {
+		dev_dbg(&pf->pdev->dev,
+			"Error: cnt_q_avail (%u) less than num_queues %d\n",

Either use parens around both numbers or don't use them on either

Quoted 36 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +36 -0
+			vsi->cnt_q_avail, ch->num_queue_pairs);
+		return -EINVAL;
+	}
+
+	/* reconfig_rss only if vsi type is MAIN_VSI */
+	if (reconfig_rss && (vsi->type == I40E_VSI_MAIN)) {
+		err = i40e_vsi_reconfig_rss(vsi, ch->num_queue_pairs);
+		if (err) {
+			dev_info(&pf->pdev->dev,
+				 "Error: unable to reconfig rss for num_queues (%u)\n",
+				 ch->num_queue_pairs);
+			return -EINVAL;
+		}
+	}
+
+	if (!i40e_setup_channel(pf, vsi, ch)) {
+		dev_info(&pf->pdev->dev, "Failed to setup channel\n");
+		return -EINVAL;
+	}
+
+	dev_info(&pf->pdev->dev,
+		 "Setup channel (id:%u) utilizing num_queues %d\n",
+		 ch->seid, ch->num_queue_pairs);
+
+	/* in case of VF, this will be main SRIOV VSI */
+	ch->parent_vsi = vsi;
+
+	/* and update main_vsi's count for queue_available to use */
+	vsi->cnt_q_avail -= ch->num_queue_pairs;
+
+	return 0;
+}
+
+/**
+ * i40e_configure_queue_channels - Add queue channel for the given TCs
+ * @vsi: VSI to be configured

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5823

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+ *
+ * Configures queue channel mapping to the given TCs
+ **/
+static int i40e_configure_queue_channels(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch;
+	int ret = 0, i;
+
+	/* Create app vsi with the TCs. Main VSI with TC0 is already set up */
+	for (i = 1; i < I40E_MAX_TRAFFIC_CLASS; i++)

There should be {}'s around this block

Quoted 132 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +78 -2
+		if (vsi->tc_config.enabled_tc & BIT(i)) {
+			ch = kzalloc(sizeof(*ch), GFP_KERNEL);
+			if (!ch) {
+				ret = -ENOMEM;
+				goto err_free;
+			}
+
+			INIT_LIST_HEAD(&ch->list);
+			ch->num_queue_pairs =
+				vsi->tc_config.tc_info[i].qcount;
+			ch->base_queue =
+				vsi->tc_config.tc_info[i].qoffset;
+
+			list_add_tail(&ch->list, &vsi->ch_list);
+
+			ret = i40e_create_queue_channel(vsi, ch);
+			if (ret) {
+				dev_err(&vsi->back->pdev->dev,
+					"Failed creating queue channel with TC%d: queues %d\n",
+					i, ch->num_queue_pairs);
+				goto err_free;
+			}
+		}
+	return ret;
+
+err_free:
+	i40e_remove_queue_channels(vsi);
+	return ret;
+}
+
+/**
   * i40e_veb_config_tc - Configure TCs for given VEB
   * @veb: given VEB
   * @enabled_tc: TC bitmap
@@ -5618,10 +6302,18 @@ static int i40e_setup_tc(struct net_device *netdev, u8 tc)
  		goto exit;
  	}
 
-	/* Unquiesce VSI */
-	i40e_unquiesce_vsi(vsi);
+	if (pf->flags & I40E_FLAG_TC_MQPRIO) {
+		ret = i40e_configure_queue_channels(vsi);
+		if (ret) {
+			netdev_info(netdev,
+				    "Failed configuring queue channels\n");
+			goto exit;
+		}
+	}
 
  exit:
+	/* Unquiesce VSI */
+	i40e_unquiesce_vsi(vsi);
  	return ret;
  }
 
@@ -7022,6 +7714,35 @@ static void i40e_fdir_teardown(struct i40e_pf *pf)
  }
 
  /**
+ * i40e_rebuild_channels - Rebuilds channel VSIs if they existed before reset
+ * @vsi: PF main vsi
+ *
+ * Rebuilds channel VSIs if they existed before reset
+ **/
+static int i40e_rebuild_channels(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+	i40e_status ret;
+
+	if (list_empty(&vsi->ch_list))
+		return 0;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		if (!ch->initialized)
+			break;
+		/* Proceed with creation of channel (VMDq2) VSI */
+		ret = i40e_add_channel(vsi->back, vsi->uplink_seid, ch);
+		if (ret) {
+			dev_info(&vsi->back->pdev->dev,
+				 "failed to rebuild channels using uplink_seid %u\n",
+				 vsi->uplink_seid);
+			return ret;
+		}
+	}
+	return 0;
+}
+
+/**
   * i40e_prep_for_reset - prep for the core to reset
   * @pf: board private structure
   * @lock_acquired: indicates whether or not the lock has been acquired
@@ -7286,6 +8007,13 @@ static void i40e_rebuild(struct i40e_pf *pf, bool reinit, bool lock_acquired)
  		}
  	}
 
+	/* PF Main VSI is rebuild by now, go ahead and rebuild channel VSIs
+	 * for this main VSI if they exist
+	 */
+	ret = i40e_rebuild_channels(pf->vsi[pf->lan_vsi]);
+	if (ret)
+		goto end_unlock;
+
  	/* Reconfigure hardware for allowing smaller MSS in the case
  	 * of TSO, so that we avoid the MDD being fired and causing
  	 * a reset in the case of small MSS+TSO.
@@ -11562,6 +12290,7 @@ static int i40e_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
  		dev_info(&pdev->dev, "setup_pf_switch failed: %d\n", err);
  		goto err_vsis;
  	}
+	INIT_LIST_HEAD(&pf->vsi[pf->lan_vsi]->ch_list);
 
  	/* Make sure flow control is set according to current settings */
  	err = i40e_set_fc(hw, &set_fc_aq_fail, true);
drivers/net/ethernet/intel/i40e/i40e_txrx.h +2 -0
index 2f848bc..a767f87 100644
@@ -423,6 +423,8 @@ struct i40e_ring {
  					 * i40e_clean_rx_ring_irq() is called
  					 * for this ring.
  					 */
+
+	struct i40e_channel *ch;
  } ____cacheline_internodealigned_in_smp;
 
  static inline bool ring_uses_build_skb(struct i40e_ring *ring)

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

#7 lore raw

PATCH v2 3/6 net: i40e: Add infrastructure for queue channel support with the TCs and queue configurations offloaded via mqprio scheduler

Quoted 309 lines · Shannon Nelson
Quoted 10 lines · Amritha Nambiar

This patch sets up the infrastructure for offloading TCs and queue configurations to the hardware by creating HW channels(VSI). A new channel is created for each of the traffic class configuration offloaded via mqprio framework except for the first TC (TC0). TC0 for the main VSI is also reconfigured as per user provided queue parameters. Queue counts that are not power-of-2 are handled by reconfiguring RSS by reprogramming LUTs using the queue count value. This patch also handles configuring the TX rings for the channels, setting up the RX queue map for channel.

Quoted 3 lines

Also, the channels so created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.

To reference an old pet-peave of mine, these channels might be how to support macvlan offloads - you might look into how these can support ndo_dfwd_add_station and ndo_dfwd_del_station.

Quoted 290 lines
---
  drivers/net/ethernet/intel/i40e/i40e.h      |   33 +
  drivers/net/ethernet/intel/i40e/i40e_main.c |  747 +++++++++++++++++++++++++++
  drivers/net/ethernet/intel/i40e/i40e_txrx.h |    2
  3 files changed, 773 insertions(+), 9 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +33 -0
index 0e53838..e259a67 100644
@@ -87,6 +87,7 @@
  #define I40E_AQ_LEN			256
  #define I40E_AQ_WORK_LIMIT		66 /* max number of VFs + a little */
  #define I40E_MAX_USER_PRIORITY		8
+#define I40E_MAX_QUEUES_PER_CH		64
  #define I40E_DEFAULT_TRAFFIC_CLASS	BIT(0)
  #define I40E_DEFAULT_MSG_ENABLE		4
  #define I40E_QUEUE_WAIT_RETRY_LIMIT	10
@@ -340,6 +341,23 @@ struct i40e_flex_pit {
  	u8 pit_index;
  };
 
+struct i40e_channel {
+	struct list_head list;
+	bool initialized;
+	u8 type;
+	u16 vsi_number; /* Assigned VSI number from AQ 'Add VSI' response */
+	u16 stat_counter_idx;
+	u16 base_queue;
+	u16 num_queue_pairs; /* Requested by user */
+	u16 seid;
+
+	u8 enabled_tc;
+	struct i40e_aqc_vsi_properties_data info;
+
+	/* track this channel belongs to which VSI */
+	struct i40e_vsi *parent_vsi;
+};
+
  /* struct that defines the Ethernet device */
  struct i40e_pf {
  	struct pci_dev *pdev;
@@ -455,6 +473,7 @@ struct i40e_pf {
  #define I40E_FLAG_CLIENT_L2_CHANGE		BIT_ULL(25)
  #define I40E_FLAG_CLIENT_RESET			BIT_ULL(26)
  #define I40E_FLAG_LINK_DOWN_ON_CLOSE_ENABLED	BIT_ULL(27)
+#define I40E_FLAG_TC_MQPRIO			BIT_ULL(28)
 
  	struct i40e_client_instance *cinst;
  	bool stat_offsets_loaded;
@@ -535,6 +554,8 @@ struct i40e_pf {
  	u32 ioremap_len;
  	u32 fd_inv;
  	u16 phy_led_val;
+
+	u16 override_q_count;
  };
 
  /**
@@ -699,6 +720,16 @@ struct i40e_vsi {
  	bool current_isup;	/* Sync 'link up' logging */
  	enum i40e_aq_link_speed current_speed;	/* Sync link speed logging */
 
+	/* channel specific fields */
+	u16 cnt_q_avail; /* num of queues available for channel usage */
+	u16 orig_rss_size;
+	u16 current_rss_size;
+
+	/* keeps track of next_base_queue to be used for channel setup */
+	atomic_t next_base_queue;
+
+	struct list_head ch_list;
+
  	void *priv;	/* client driver data reference. */
 
  	/* VSI specific handlers */
@@ -1006,4 +1037,6 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi)
  {
  	return !!vsi->xdp_prog;
  }
+
+int i40e_create_queue_channel(struct i40e_vsi *vsi, struct i40e_channel *ch);
  #endif /* _I40E_H_ */
drivers/net/ethernet/intel/i40e/i40e_main.c +167 -7
index 9392c5a..64769cb 100644
@@ -2887,7 +2887,7 @@ static void i40e_config_xps_tx_ring(struct i40e_ring *ring)
  {
  	struct i40e_vsi *vsi = ring->vsi;
 
-	if (!ring->q_vector || !ring->netdev)
+	if (!ring->q_vector || !ring->netdev || ring->ch)
  		return;
 
  	if ((vsi->tc_config.numtc <= 1) &&
@@ -2954,7 +2954,14 @@ static int i40e_configure_tx_ring(struct i40e_ring *ring)
  	 * initialization. This has to be done regardless of
  	 * DCB as by default everything is mapped to TC0.
  	 */
-	tx_ctx.rdylist = le16_to_cpu(vsi->info.qs_handle[ring->dcb_tc]);
+
+	if (ring->ch)
+		tx_ctx.rdylist =
+			le16_to_cpu(ring->ch->info.qs_handle[ring->dcb_tc]);
+
+	else
+		tx_ctx.rdylist = le16_to_cpu(vsi->info.qs_handle[ring->dcb_tc]);
+
  	tx_ctx.rdylist_act = 0;
 
  	/* clear the context in the HMC */
@@ -2976,12 +2983,23 @@ static int i40e_configure_tx_ring(struct i40e_ring *ring)
  	}
 
  	/* Now associate this queue with this PCI function */
-	if (vsi->type == I40E_VSI_VMDQ2) {
-		qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
-		qtx_ctl |= ((vsi->id) << I40E_QTX_CTL_VFVM_INDX_SHIFT) &
-			   I40E_QTX_CTL_VFVM_INDX_MASK;
+	if (ring->ch) {
+		if (ring->ch->type == I40E_VSI_VMDQ2)
+			qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
+		else
+			return -EINVAL;
+
+		qtx_ctl |= (ring->ch->vsi_number <<
+			    I40E_QTX_CTL_VFVM_INDX_SHIFT) &
+			    I40E_QTX_CTL_VFVM_INDX_MASK;
  	} else {
-		qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
+		if (vsi->type == I40E_VSI_VMDQ2) {
+			qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
+			qtx_ctl |= ((vsi->id) << I40E_QTX_CTL_VFVM_INDX_SHIFT) &
+				    I40E_QTX_CTL_VFVM_INDX_MASK;
+		} else {
+			qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
+		}
  	}
 
  	qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
@@ -5168,6 +5186,672 @@ static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  }
 
  /**
+ * i40e_remove_queue_channels - Remove queue channels for the TCs
+ * @vsi: VSI to be configured
+ *
+ * Remove queue channels for the TCs
+ **/
+static void i40e_remove_queue_channels(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+	int ret, i;
+
+	/* Reset rss size that was stored when reconfiguring rss for
+	 * channel VSIs with non-power-of-2 queue count.
+	 */
+	vsi->current_rss_size = 0;
+
+	/* perform cleanup for channels if they exist */
+	if (list_empty(&vsi->ch_list))
+		return;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		struct i40e_vsi *p_vsi;
+
+		list_del(&ch->list);
+		p_vsi = ch->parent_vsi;
+		if (!p_vsi || !ch->initialized) {
+			kfree(ch);
+			continue;
+		}
+		/* Reset queue contexts */
+		for (i = 0; i < ch->num_queue_pairs; i++) {
+			struct i40e_ring *tx_ring, *rx_ring;
+			u16 pf_q;
+
+			pf_q = ch->base_queue + i;
+			tx_ring = vsi->tx_rings[pf_q];
+			tx_ring->ch = NULL;
+
+			rx_ring = vsi->rx_rings[pf_q];
+			rx_ring->ch = NULL;
+		}
+
+		/* delete VSI from FW */
+		ret = i40e_aq_delete_element(&vsi->back->hw, ch->seid,
+					     NULL);
+		if (ret)
+			dev_err(&vsi->back->pdev->dev,
+				"unable to remove channel (%d) for parent VSI(%d)\n",
+				ch->seid, p_vsi->seid);
+		kfree(ch);
+	}
+	INIT_LIST_HEAD(&vsi->ch_list);
+}
+
+/**
+ * i40e_is_any_channel - channel exist or not
+ * @vsi: ptr to VSI to which channels are associated with
+ *
+ * Returns true or false if channel(s) exist for associated VSI or not
+ **/
+static bool i40e_is_any_channel(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		if (ch->initialized)
+			return true;
+	}
+
+	return false;
+}
+
+/**
+ * i40e_get_max_queues_for_channel
+ * @vsi: ptr to VSI to which channels are associated with
+ *
+ * Helper function which returns max value among the queue counts set on the
+ * channels/TCs created.
+ **/
+static int i40e_get_max_queues_for_channel(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+	int max = 0;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		if (!ch->initialized)
+			continue;
+		if (ch->num_queue_pairs > max)
+			max = ch->num_queue_pairs;
+	}
+
+	return max;
+}
+
+/**
+ * i40e_validate_num_queues - validate num_queues w.r.t channel
+ * @pf: ptr to PF device
+ * @num_queues: number of queues
+ * @vsi: the parent VSI
+ * @reconfig_rss: indicates should the RSS be reconfigured or not
+ *
+ * This function validates number of queues in the context of new channel
+ * which is being established and determines if RSS should be reconfigured
+ * or not for parent VSI.
+ **/
+static int i40e_validate_num_queues(struct i40e_pf *pf, int num_queues,
+				    struct i40e_vsi *vsi, bool *reconfig_rss)
+{
+	int max_ch_queues;
+
+	if (!reconfig_rss)
+		return -EINVAL;
+
+	*reconfig_rss = false;
+
+	if (num_queues > I40E_MAX_QUEUES_PER_CH) {
+		dev_err(&pf->pdev->dev,
+			"Failed to create VMDq VSI. User requested num_queues (%d) > I40E_MAX_QUEUES_PER_VSI (%u)\n",
+			num_queues, I40E_MAX_QUEUES_PER_CH);
+		return -EINVAL;
+	}
+
+	if (vsi->current_rss_size) {
+		if (num_queues > vsi->current_rss_size) {
+			dev_dbg(&pf->pdev->dev,
+				"Error: num_queues (%d) > vsi's current_size(%d)\n",
+				num_queues, vsi->current_rss_size);
+			return -EINVAL;
+		} else if ((num_queues < vsi->current_rss_size) &&
+			   (!is_power_of_2(num_queues))) {
+			dev_dbg(&pf->pdev->dev,
+				"Error: num_queues (%d) < vsi's current_size(%d), but not power of 2\n",
+				num_queues, vsi->current_rss_size);
+			return -EINVAL;
+		}
+	}
+
+	if (!is_power_of_2(num_queues)) {
+		/* Find the max num_queues configures for channel if channel
+		 * exist.
+		 * if channel exist, then enforce 'num_queues' to be more than
+		 * max ever num_queues configured for channel.
+		 */
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5335

Quoted 5 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +5 -0
+		max_ch_queues = i40e_get_max_queues_for_channel(vsi);
+		if (num_queues < max_ch_queues) {
+			dev_dbg(&pf->pdev->dev,
+				"Error: num_queues (%d) > main_vsi's current_size(%d)\n",
+				num_queues, vsi->current_rss_size);

This doesn't look like it matches the if expression - should the expression be '>' rather than '<'?

Will fix in v3.

Quoted 81 lines
Quoted 81 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +80 -0
+			return -EINVAL;
+		}
+		*reconfig_rss = true;
+	}
+
+	return 0;
+}
+
+/**
+ * i40e_vsi_reconfig_rss - reconfig RSS based on specified rss_size
+ * @vsi: the VSI being setup
+ * @rss_size: size of RSS, accordingly LUT gets reprogrammed
+ *
+ * This function reconfigures RSS by reprogramming LUTs using 'rss_size'
+ **/
+static int i40e_vsi_reconfig_rss(struct i40e_vsi *vsi, u16 rss_size)
+{
+	struct i40e_pf *pf = vsi->back;
+	u8 seed[I40E_HKEY_ARRAY_SIZE];
+	struct i40e_hw *hw = &pf->hw;
+	int local_rss_size;
+	u8 *lut;
+	int ret;
+
+	if (!vsi->rss_size)
+		return -EINVAL;
+
+	if (rss_size > vsi->rss_size)
+		return -EINVAL;
+
+	local_rss_size = min_t(int, vsi->rss_size, rss_size);
+	lut = kzalloc(vsi->rss_table_size, GFP_KERNEL);
+	if (!lut)
+		return -ENOMEM;
+
+	/* Ignoring user configured lut if there is one */
+	i40e_fill_rss_lut(pf, lut, vsi->rss_table_size, local_rss_size);
+
+	/* Use user configured hash key if there is one, otherwise
+	 * use default.
+	 */
+	if (vsi->rss_hkey_user)
+		memcpy(seed, vsi->rss_hkey_user, I40E_HKEY_ARRAY_SIZE);
+	else
+		netdev_rss_key_fill((void *)seed, I40E_HKEY_ARRAY_SIZE);
+
+	ret = i40e_config_rss(vsi, seed, lut, vsi->rss_table_size);
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "Cannot set RSS lut, err %s aq_err %s\n",
+			 i40e_stat_str(hw, ret),
+			 i40e_aq_str(hw, hw->aq.asq_last_status));
+		kfree(lut);
+		return ret;
+	}
+	kfree(lut);
+
+	/* Do the update w.r.t. storing rss_size */
+	if (!vsi->orig_rss_size)
+		vsi->orig_rss_size = vsi->rss_size;
+	vsi->current_rss_size = local_rss_size;
+
+	return ret;
+}
+
+/**
+ * i40e_channel_setup_queue_map - Setup a channel queue map based on enabled_tc
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ctxt: VSI context structure
+ * @enabled_tc: Enabled TCs bitmap
+ * @ch: ptr to channel structure
+ *
+ * Setup queue map based on enabled_tc for specific channel
+ **/
+static void i40e_channel_setup_queue_map(struct i40e_pf *pf,
+					 struct i40e_vsi_context *ctxt,
+					 u8 enabled_tc, struct i40e_channel *ch)
+{
+	u16 qcount, num_tc_qps, qmap;
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5420

Quoted 5 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +5 -0
+	int pow, num_qps;
+	u16 sections = 0;
+	/* At least TC0 is enabled in case of non-DCB case, non-MQPRIO */
+	u16 numtc = 1;
+	u8 offset = 0;

The enabled_tc parameter and the numtc variable are useless here, I assume they'll be useful in a later patch?

Will fix in v3.

Quoted 226 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5431

Quoted 12 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +11 -0
+
+	sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID;
+	sections |= I40E_AQ_VSI_PROP_SCHED_VALID;
+
+	if (pf->flags & I40E_FLAG_MSIX_ENABLED) {
+		qcount = min_t(int, ch->num_queue_pairs,
+			       pf->num_lan_msix);
+		ch->num_queue_pairs = qcount;
+	} else {
+		qcount = 1;
+	}

If no MSIX, we probably shouldn't even be here.

Quoted 211 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +211 -0
+
+	/* find num of qps per traffic class */
+	num_tc_qps = qcount / numtc;
+	num_tc_qps = min_t(int, num_tc_qps, i40e_pf_get_max_q_per_tc(pf));
+	num_qps = qcount;
+
+	/* find the next higher power-of-2 of num queue pairs */
+	pow = ilog2(num_qps);
+	if (!is_power_of_2(num_qps))
+		pow++;
+
+	qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) |
+		(pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT);
+
+	/* Setup queue TC[0].qmap for given VSI context */
+	ctxt->info.tc_mapping[0] = cpu_to_le16(qmap);
+
+	ctxt->info.up_enable_bits = enabled_tc;
+	ctxt->info.mapping_flags |= cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG);
+	ctxt->info.queue_mapping[0] = cpu_to_le16(ch->base_queue);
+	ctxt->info.valid_sections |= cpu_to_le16(sections);
+}
+
+/**
+ * i40e_add_channel - add a channel by adding VSI
+ * @pf: ptr to PF device
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @ch: ptr to channel structure
+ *
+ * Add a channel (VSI) using add_vsi and queue_map
+ **/
+static int i40e_add_channel(struct i40e_pf *pf, u16 uplink_seid,
+			    struct i40e_channel *ch)
+{
+	struct i40e_hw *hw = &pf->hw;
+	struct i40e_vsi_context ctxt;
+	u8 enabled_tc = 0x1; /* TC0 enabled */
+	int ret;
+
+	if (ch->type != I40E_VSI_VMDQ2) {
+		dev_info(&pf->pdev->dev,
+			 "add new vsi failed, ch->type %d\n", ch->type);
+		return -EINVAL;
+	}
+
+	memset(&ctxt, 0, sizeof(ctxt));
+	ctxt.pf_num = hw->pf_id;
+	ctxt.vf_num = 0;
+	ctxt.uplink_seid = uplink_seid;
+	ctxt.connection_type = I40E_AQ_VSI_CONN_TYPE_NORMAL;
+	if (ch->type == I40E_VSI_VMDQ2)
+		ctxt.flags = I40E_AQ_VSI_TYPE_VMDQ2;
+
+	if (pf->flags & I40E_FLAG_VEB_MODE_ENABLED) {
+		ctxt.info.valid_sections |=
+		     cpu_to_le16(I40E_AQ_VSI_PROP_SWITCH_VALID);
+		ctxt.info.switch_id =
+		   cpu_to_le16(I40E_AQ_VSI_SW_ID_FLAG_ALLOW_LB);
+	}
+
+	/* Set queue map for a given VSI context */
+	i40e_channel_setup_queue_map(pf, &ctxt, enabled_tc, ch);
+
+	/* Now time to create VSI */
+	ret = i40e_aq_add_vsi(hw, &ctxt, NULL);
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "add new vsi failed, err %s aq_err %s\n",
+			 i40e_stat_str(&pf->hw, ret),
+			 i40e_aq_str(&pf->hw,
+				     pf->hw.aq.asq_last_status));
+		return -ENOENT;
+	}
+
+	/* Success, update channel */
+	ch->enabled_tc = enabled_tc;
+	ch->seid = ctxt.seid;
+	ch->vsi_number = ctxt.vsi_number;
+	ch->stat_counter_idx = cpu_to_le16(ctxt.info.stat_counter_idx);
+
+	/* copy just the sections touched not the entire info
+	 * since not all sections are valid as returned by
+	 * update vsi params
+	 */
+	ch->info.mapping_flags = ctxt.info.mapping_flags;
+	memcpy(&ch->info.queue_mapping,
+	       &ctxt.info.queue_mapping, sizeof(ctxt.info.queue_mapping));
+	memcpy(&ch->info.tc_mapping, ctxt.info.tc_mapping,
+	       sizeof(ctxt.info.tc_mapping));
+
+	return 0;
+}
+
+static int i40e_channel_config_bw(struct i40e_vsi *vsi, struct i40e_channel *ch,
+				  u8 *bw_share)
+{
+	struct i40e_aqc_configure_vsi_tc_bw_data bw_data;
+	i40e_status ret;
+	int i;
+
+	bw_data.tc_valid_bits = ch->enabled_tc;
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
+		bw_data.tc_bw_credits[i] = bw_share[i];
+
+	ret = i40e_aq_config_vsi_tc_bw(&vsi->back->hw, ch->seid,
+				       &bw_data, NULL);
+	if (ret) {
+		dev_info(&vsi->back->pdev->dev,
+			 "Config VSI BW allocation per TC failed, aq_err: %d for new_vsi->seid %u\n",
+			 vsi->back->hw.aq.asq_last_status, ch->seid);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
+		ch->info.qs_handle[i] = bw_data.qs_handles[i];
+
+	return 0;
+}
+
+/**
+ * i40e_channel_config_tx_ring - config TX ring associated with new channel
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ch: ptr to channel structure
+ *
+ * Configure TX rings associated with channel (VSI) since queues are being
+ * from parent VSI.
+ **/
+static int i40e_channel_config_tx_ring(struct i40e_pf *pf,
+				       struct i40e_vsi *vsi,
+				       struct i40e_channel *ch)
+{
+	i40e_status ret;
+	int i;
+	u8 bw_share[I40E_MAX_TRAFFIC_CLASS] = {0};
+
+	/* Enable ETS TCs with equal BW Share for now across all VSIs */
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+		if (ch->enabled_tc & BIT(i))
+			bw_share[i] = 1;
+	}
+
+	/* configure BW for new VSI */
+	ret = i40e_channel_config_bw(vsi, ch, bw_share);
+	if (ret) {
+		dev_info(&vsi->back->pdev->dev,
+			 "Failed configuring TC map %d for channel (seid %u)\n",
+			 ch->enabled_tc, ch->seid);
+		return ret;
+	}
+
+	for (i = 0; i < ch->num_queue_pairs; i++) {
+		struct i40e_ring *tx_ring, *rx_ring;
+		u16 pf_q;
+
+		pf_q = ch->base_queue + i;
+
+		/* Get to TX ring ptr of main VSI, for re-setup TX queue
+		 * context
+		 */
+		tx_ring = vsi->tx_rings[pf_q];
+		tx_ring->ch = ch;
+
+		/* Get the RX ring ptr */
+		rx_ring = vsi->rx_rings[pf_q];
+		rx_ring->ch = ch;
+	}
+
+	return 0;
+}
+
+/**
+ * i40e_setup_hw_channel - setup new channel
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ch: ptr to channel structure
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @type: type of channel to be created (VMDq2/VF)
+ *
+ * Setup new channel (VSI) based on specified type (VMDq2/VF)
+ * and configures TX rings accordingly
+ **/
+static inline int i40e_setup_hw_channel(struct i40e_pf *pf,
+					struct i40e_vsi *vsi,
+					struct i40e_channel *ch,
+					u16 uplink_seid, u8 type)
+{
+	struct i40e_q_vector *q_vector;
+	int base_queue = 0;
+	int ret, i;
+
+	ch->initialized = false;
+	ch->base_queue = atomic_read(&vsi->next_base_queue);
+	ch->type = type;
+
+	/* Proceed with creation of channel (VMDq2) VSI */
+	ret = i40e_add_channel(pf, uplink_seid, ch);
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "failed to add_channel using uplink_seid %u\n",
+			 uplink_seid);
+		return ret;
+	}
+
+	/* Mark the successful creation of channel */
+	ch->initialized = true;
+
+	/* Mark q_vectors indicating that they are part of newly created
+	 * channel (VSI)
+	 */
+	base_queue = ch->base_queue;
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5647

Quoted 5 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +5 -0
+	for (i = 0; i < ch->num_queue_pairs; i++) {
+		q_vector = vsi->tx_rings[base_queue + i]->q_vector;
+
+		if (!q_vector)
+			continue;

This seems like a fairly useless for-loop. Perhaps a later patch adds something here?

Will fix in v3.

Quoted 11 lines
Quoted 11 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+	}
+
+	/* Reconfigure TX queues using QTX_CTL register */
+	ret = i40e_channel_config_tx_ring(pf, vsi, ch);
+	if (ret) {
+		dev_info(&pf->pdev->dev,
+			 "failed to configure TX rings for channel %u\n",
+			 ch->seid);
+		return ret;
+	}
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5661

Quoted 4 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +4 -0
+
+	/* update 'next_base_queue' */
+	atomic_set(&vsi->next_base_queue,
+		   atomic_read(&vsi->next_base_queue) + ch->num_queue_pairs);

Is there any possibility that this can get changed by a separate thread in between getting the base_queue and here? If not, why bother with atomics? If so, shouldn't there be a lock between here and there?

Will fix in v3.

Quoted 19 lines
Quoted 19 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +18 -0
+
+	dev_dbg(&pf->pdev->dev,
+		"Added channel: vsi_seid %u, vsi_number %u, stat_counter_idx %u, num_queue_pairs %u, pf->next_base_queue %d\n",
+		ch->seid, ch->vsi_number, ch->stat_counter_idx,
+		ch->num_queue_pairs,
+		atomic_read(&vsi->next_base_queue));
+
+	return ret;
+}
+
+/**
+ * i40e_setup_channel - setup new channel using uplink element
+ * @pf: ptr to PF device
+ * @type: type of channel to be created (VMDq2/VF)
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @ch: ptr to channel structure
+ *
+ * Setup new channel (VSI) based on specified type (VMDq2/VF)
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5685

Quoted 6 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +6 -0
+ * and uplink switching element (uplink_seid)
+ **/
+static bool i40e_setup_channel(struct i40e_pf *pf, struct i40e_vsi *vsi,
+			       struct i40e_channel *ch)
+{
+	u16 seid = vsi->seid;

Why set this when it gets set differently in a couple of lines below?

Will fix in v3.

Quoted 10 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5692

Quoted 8 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +7 -0
+	u8 vsi_type;
+	int ret;
+
+	if (vsi->type == I40E_VSI_MAIN) {
+		vsi_type = I40E_VSI_VMDQ2;
+	} else {
+		dev_err(&pf->pdev->dev, "unsupported vsi type(%d) of parent vsi\n",

Might be better worded as "unsupported parent vsi type (%d)"

Will fix in v3.

Quoted 79 lines
Quoted 79 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +78 -0
+			vsi->type);
+		return false;
+	} > +
+	/* underlying switching element */
+	seid = pf->vsi[pf->lan_vsi]->uplink_seid;
+
+	/* create channel (VSI), configure TX rings */
+	ret = i40e_setup_hw_channel(pf, vsi, ch, seid, vsi_type);
+	if (ret) {
+		dev_err(&pf->pdev->dev, "failed to setup hw_channel\n");
+		return false;
+	}
+
+	return ch->initialized ? true : false;
+}
+
+/**
+ * i40e_create_queue_channel - function to create channel
+ * @vsi: VSI to be configured
+ * @ch: ptr to channel (it contains channel specific params)
+ *
+ * This function creates channel (VSI) using num_queues specified by user,
+ * reconfigs RSS if needed.
+ **/
+int i40e_create_queue_channel(struct i40e_vsi *vsi,
+			      struct i40e_channel *ch)
+{
+	struct i40e_pf *pf = vsi->back;
+	bool reconfig_rss;
+	int err;
+
+	if (!ch)
+		return -EINVAL;
+
+	if (!ch->num_queue_pairs) {
+		dev_err(&pf->pdev->dev, "Invalid num_queues requested: %d\n",
+			ch->num_queue_pairs);
+		return -EINVAL;
+	}
+
+	/* validate user requested num_queues for channel */
+	err = i40e_validate_num_queues(pf, ch->num_queue_pairs, vsi,
+				       &reconfig_rss);
+	if (err) {
+		dev_info(&pf->pdev->dev, "Failed to validate num_queues (%d)\n",
+			 ch->num_queue_pairs);
+		return -EINVAL;
+	}
+
+	/* By default we are in VEPA mode, if this is the first VF/VMDq
+	 * VSI to be added switch to VEB mode.
+	 */
+	if ((!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) ||
+	    (!i40e_is_any_channel(vsi))) {
+		if (!is_power_of_2(vsi->tc_config.tc_info[0].qcount)) {
+			dev_dbg(&pf->pdev->dev,
+				"Failed to create channel. Override queues (%u) not power of 2\n",
+				vsi->tc_config.tc_info[0].qcount);
+			return -EINVAL;
+		}
+
+		if (!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) {
+			pf->flags |= I40E_FLAG_VEB_MODE_ENABLED;
+
+			if (vsi->type == I40E_VSI_MAIN) {
+				if (pf->flags & I40E_FLAG_TC_MQPRIO)
+					i40e_do_reset(pf, I40E_PF_RESET_FLAG,
+						      true);
+				else
+					i40e_do_reset_safe(pf,
+							   I40E_PF_RESET_FLAG);
+			}
+		}
+		/* now onwards for main VSI, number of queues will be value
+		 * of TC0's queue count
+		 */
+	}
+
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5777

Quoted 6 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +6 -0
+	/* By this time, vsi->cnt_q_avail shall be set to non-zero and
+	 * it should be more than num_queues
+	 */
+	if (!vsi->cnt_q_avail || (vsi->cnt_q_avail < ch->num_queue_pairs)) {
+		dev_dbg(&pf->pdev->dev,
+			"Error: cnt_q_avail (%u) less than num_queues %d\n",

Either use parens around both numbers or don't use them on either

Will fix in v3.

Quoted 41 lines
Quoted 41 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +40 -0
+			vsi->cnt_q_avail, ch->num_queue_pairs);
+		return -EINVAL;
+	}
+
+	/* reconfig_rss only if vsi type is MAIN_VSI */
+	if (reconfig_rss && (vsi->type == I40E_VSI_MAIN)) {
+		err = i40e_vsi_reconfig_rss(vsi, ch->num_queue_pairs);
+		if (err) {
+			dev_info(&pf->pdev->dev,
+				 "Error: unable to reconfig rss for num_queues (%u)\n",
+				 ch->num_queue_pairs);
+			return -EINVAL;
+		}
+	}
+
+	if (!i40e_setup_channel(pf, vsi, ch)) {
+		dev_info(&pf->pdev->dev, "Failed to setup channel\n");
+		return -EINVAL;
+	}
+
+	dev_info(&pf->pdev->dev,
+		 "Setup channel (id:%u) utilizing num_queues %d\n",
+		 ch->seid, ch->num_queue_pairs);
+
+	/* in case of VF, this will be main SRIOV VSI */
+	ch->parent_vsi = vsi;
+
+	/* and update main_vsi's count for queue_available to use */
+	vsi->cnt_q_avail -= ch->num_queue_pairs;
+
+	return 0;
+}
+
+/**
+ * i40e_configure_queue_channels - Add queue channel for the given TCs
+ * @vsi: VSI to be configured
+ *
+ * Configures queue channel mapping to the given TCs
+ **/
+static int i40e_configure_queue_channels(struct i40e_vsi *vsi)
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5823

Quoted 6 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +6 -0
+{
+	struct i40e_channel *ch;
+	int ret = 0, i;
+
+	/* Create app vsi with the TCs. Main VSI with TC0 is already set up */
+	for (i = 1; i < I40E_MAX_TRAFFIC_CLASS; i++)

There should be {}'s around this block

Will fix in v3.

Quoted 133 lines
Quoted 133 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +78 -2
+		if (vsi->tc_config.enabled_tc & BIT(i)) {
+			ch = kzalloc(sizeof(*ch), GFP_KERNEL);
+			if (!ch) {
+				ret = -ENOMEM;
+				goto err_free;
+			}
+
+			INIT_LIST_HEAD(&ch->list);
+			ch->num_queue_pairs =
+				vsi->tc_config.tc_info[i].qcount;
+			ch->base_queue =
+				vsi->tc_config.tc_info[i].qoffset;
+
+			list_add_tail(&ch->list, &vsi->ch_list);
+
+			ret = i40e_create_queue_channel(vsi, ch);
+			if (ret) {
+				dev_err(&vsi->back->pdev->dev,
+					"Failed creating queue channel with TC%d: queues %d\n",
+					i, ch->num_queue_pairs);
+				goto err_free;
+			}
+		}
+	return ret;
+
+err_free:
+	i40e_remove_queue_channels(vsi);
+	return ret;
+}
+
+/**
   * i40e_veb_config_tc - Configure TCs for given VEB
   * @veb: given VEB
   * @enabled_tc: TC bitmap
@@ -5618,10 +6302,18 @@ static int i40e_setup_tc(struct net_device *netdev, u8 tc)
  		goto exit;
  	}
 
-	/* Unquiesce VSI */
-	i40e_unquiesce_vsi(vsi);
+	if (pf->flags & I40E_FLAG_TC_MQPRIO) {
+		ret = i40e_configure_queue_channels(vsi);
+		if (ret) {
+			netdev_info(netdev,
+				    "Failed configuring queue channels\n");
+			goto exit;
+		}
+	}
 
  exit:
+	/* Unquiesce VSI */
+	i40e_unquiesce_vsi(vsi);
  	return ret;
  }
 
@@ -7022,6 +7714,35 @@ static void i40e_fdir_teardown(struct i40e_pf *pf)
  }
 
  /**
+ * i40e_rebuild_channels - Rebuilds channel VSIs if they existed before reset
+ * @vsi: PF main vsi
+ *
+ * Rebuilds channel VSIs if they existed before reset
+ **/
+static int i40e_rebuild_channels(struct i40e_vsi *vsi)
+{
+	struct i40e_channel *ch, *ch_tmp;
+	i40e_status ret;
+
+	if (list_empty(&vsi->ch_list))
+		return 0;
+
+	list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+		if (!ch->initialized)
+			break;
+		/* Proceed with creation of channel (VMDq2) VSI */
+		ret = i40e_add_channel(vsi->back, vsi->uplink_seid, ch);
+		if (ret) {
+			dev_info(&vsi->back->pdev->dev,
+				 "failed to rebuild channels using uplink_seid %u\n",
+				 vsi->uplink_seid);
+			return ret;
+		}
+	}
+	return 0;
+}
+
+/**
   * i40e_prep_for_reset - prep for the core to reset
   * @pf: board private structure
   * @lock_acquired: indicates whether or not the lock has been acquired
@@ -7286,6 +8007,13 @@ static void i40e_rebuild(struct i40e_pf *pf, bool reinit, bool lock_acquired)
  		}
  	}
 
+	/* PF Main VSI is rebuild by now, go ahead and rebuild channel VSIs
+	 * for this main VSI if they exist
+	 */
+	ret = i40e_rebuild_channels(pf->vsi[pf->lan_vsi]);
+	if (ret)
+		goto end_unlock;
+
  	/* Reconfigure hardware for allowing smaller MSS in the case
  	 * of TSO, so that we avoid the MDD being fired and causing
  	 * a reset in the case of small MSS+TSO.
@@ -11562,6 +12290,7 @@ static int i40e_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
  		dev_info(&pdev->dev, "setup_pf_switch failed: %d\n", err);
  		goto err_vsis;
  	}
+	INIT_LIST_HEAD(&pf->vsi[pf->lan_vsi]->ch_list);
 
  	/* Make sure flow control is set according to current settings */
  	err = i40e_set_fc(hw, &set_fc_aq_fail, true);
drivers/net/ethernet/intel/i40e/i40e_txrx.h +2 -0
index 2f848bc..a767f87 100644
@@ -423,6 +423,8 @@ struct i40e_ring {
  					 * i40e_clean_rx_ring_irq() is called
  					 * for this ring.
  					 */
+
+	struct i40e_channel *ch;
  } ____cacheline_internodealigned_in_smp;
 
  static inline bool ring_uses_build_skb(struct i40e_ring *ring)

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

#8 lore raw

PATCH v2 4/6 net: i40e: Enable mqprio full offload mode in the i40e driver for configuring TCs and queue mapping

In reply to Amritha Nambiar

The i40e driver is modified to enable the new mqprio hardware offload mode and factor the TCs and queue configuration by creating channel VSIs. In this mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic classes when the 'hw' option is set to 2.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 4\
  map 0 0 0 0 1 2 2 3 queues 2 at 0 2 at 2 1 at 4 1 at 5 hw 2
# tc qdisc show dev eth0

qdisc mqprio 8038: root tc 4 map 0 0 0 0 1 2 2 3 0 0 0 0 0 0 0 0

             queues:(0:1) (2:3) (4:4) (5:5)

The HW channels created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.

# tc qdisc del dev eth0 root

This patch also disables setting up channels via ethtool (ethtool -L) when the TCs are configured using mqprio scheduler.

The patch also limits setting ethtool Rx flow hash indirection (ethtool -X eth0 equal N) to max queues configured via mqprio. The Rx flow hash indirection input through ethtool should be validated so that it is within in the queue range configured via tc/mqprio. The bound checking is achieved by reporting the current rss size to the kernel when queues are configured via mqprio.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 4\
  map 0 0 0 1 0 2 3 0 queues 2 at 0 4 at 2 8 at 6 11 at 14 hw 2
# ethtool -X eth0 equal 12
Cannot set RX flow hash configuration: Invalid argument

v2: Clean up __i40e_setup_tc() and i40e_setup_tc() to work with Jiri's changes to the ndo_setup_tc interface which now takes a type and the type_data for the offload. No need to disable ATR in MQPRIO mode.

---
 drivers/net/ethernet/intel/i40e/i40e.h         |    3 
 drivers/net/ethernet/intel/i40e/i40e_ethtool.c |    8 
 drivers/net/ethernet/intel/i40e/i40e_main.c    |  488 ++++++++++++++++++------
 3 files changed, 384 insertions(+), 115 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +3 -0
index e259a67..2565280 100644
@@ -54,6 +54,7 @@
 #include <linux/clocksource.h>
 #include <linux/net_tstamp.h>
 #include <linux/ptp_clock_kernel.h>
+#include <net/pkt_cls.h>
 #include "i40e_type.h"
 #include "i40e_prototype.h"
 #include "i40e_client.h"
@@ -699,6 +700,7 @@ struct i40e_vsi {
 	enum i40e_vsi_type type;  /* VSI type, e.g., LAN, FCoE, etc */
 	s16 vf_id;		/* Virtual function ID for SRIOV VSIs */
 
+	struct tc_mqprio_qopt_offload mqprio_qopt; /* queue parameters */
 	struct i40e_tc_configuration tc_config;
 	struct i40e_aqc_vsi_properties_data info;
 
@@ -724,6 +726,7 @@ struct i40e_vsi {
 	u16 cnt_q_avail; /* num of queues available for channel usage */
 	u16 orig_rss_size;
 	u16 current_rss_size;
+	bool reconfig_rss;
 
 	/* keeps track of next_base_queue to be used for channel setup */
 	atomic_t next_base_queue;
drivers/net/ethernet/intel/i40e/i40e_ethtool.c +7 -1
index b273c24..326fc18 100644
@@ -2647,7 +2647,7 @@ static int i40e_get_rxnfc(struct net_device *netdev, struct ethtool_rxnfc *cmd,
 
 	switch (cmd->cmd) {
 	case ETHTOOL_GRXRINGS:
-		cmd->data = vsi->num_queue_pairs;
+		cmd->data = vsi->rss_size;
 		ret = 0;
 		break;
 	case ETHTOOL_GRXFH:
@@ -3892,6 +3892,12 @@ static int i40e_set_channels(struct net_device *dev,
 	if (vsi->type != I40E_VSI_MAIN)
 		return -EINVAL;
 
+	/* We do not support setting channels via ethtool when TCs are
+	 * configured through mqprio
+	 */
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return -EINVAL;
+
 	/* verify they are not requesting separate vectors */
 	if (!count || ch->rx_count || ch->tx_count)
 		return -EINVAL;
drivers/net/ethernet/intel/i40e/i40e_main.c +374 -114

619 lines not on this page · raw message

#9 lore raw

PATCH v2 4/6 net: i40e: Enable mqprio full offload mode in the i40e driver for configuring TCs and queue mapping

Quoted 182 lines · Amritha Nambiar

The i40e driver is modified to enable the new mqprio hardware offload mode and factor the TCs and queue configuration by creating channel VSIs. In this mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic classes when the 'hw' option is set to 2.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 4\
   map 0 0 0 0 1 2 2 3 queues 2 at 0 2 at 2 1 at 4 1 at 5 hw 2
# tc qdisc show dev eth0

qdisc mqprio 8038: root tc 4 map 0 0 0 0 1 2 2 3 0 0 0 0 0 0 0 0

              queues:(0:1) (2:3) (4:4) (5:5)

The HW channels created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.

# tc qdisc del dev eth0 root

This patch also disables setting up channels via ethtool (ethtool -L) when the TCs are configured using mqprio scheduler.

The patch also limits setting ethtool Rx flow hash indirection (ethtool -X eth0 equal N) to max queues configured via mqprio. The Rx flow hash indirection input through ethtool should be validated so that it is within in the queue range configured via tc/mqprio. The bound checking is achieved by reporting the current rss size to the kernel when queues are configured via mqprio.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 4\
   map 0 0 0 1 0 2 3 0 queues 2 at 0 4 at 2 8 at 6 11 at 14 hw 2
# ethtool -X eth0 equal 12
Cannot set RX flow hash configuration: Invalid argument

v2: Clean up __i40e_setup_tc() and i40e_setup_tc() to work with Jiri's changes to the ndo_setup_tc interface which now takes a type and the type_data for the offload. No need to disable ATR in MQPRIO mode.

---
  drivers/net/ethernet/intel/i40e/i40e.h         |    3
  drivers/net/ethernet/intel/i40e/i40e_ethtool.c |    8
  drivers/net/ethernet/intel/i40e/i40e_main.c    |  488 ++++++++++++++++++------
  3 files changed, 384 insertions(+), 115 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +3 -0
index e259a67..2565280 100644
@@ -54,6 +54,7 @@
  #include <linux/clocksource.h>
  #include <linux/net_tstamp.h>
  #include <linux/ptp_clock_kernel.h>
+#include <net/pkt_cls.h>
  #include "i40e_type.h"
  #include "i40e_prototype.h"
  #include "i40e_client.h"
@@ -699,6 +700,7 @@ struct i40e_vsi {
  	enum i40e_vsi_type type;  /* VSI type, e.g., LAN, FCoE, etc */
  	s16 vf_id;		/* Virtual function ID for SRIOV VSIs */
 
+	struct tc_mqprio_qopt_offload mqprio_qopt; /* queue parameters */
  	struct i40e_tc_configuration tc_config;
  	struct i40e_aqc_vsi_properties_data info;
 
@@ -724,6 +726,7 @@ struct i40e_vsi {
  	u16 cnt_q_avail; /* num of queues available for channel usage */
  	u16 orig_rss_size;
  	u16 current_rss_size;
+	bool reconfig_rss;
 
  	/* keeps track of next_base_queue to be used for channel setup */
  	atomic_t next_base_queue;
drivers/net/ethernet/intel/i40e/i40e_ethtool.c +7 -1
index b273c24..326fc18 100644
@@ -2647,7 +2647,7 @@ static int i40e_get_rxnfc(struct net_device *netdev, struct ethtool_rxnfc *cmd,
 
  	switch (cmd->cmd) {
  	case ETHTOOL_GRXRINGS:
-		cmd->data = vsi->num_queue_pairs;
+		cmd->data = vsi->rss_size;
  		ret = 0;
  		break;
  	case ETHTOOL_GRXFH:
@@ -3892,6 +3892,12 @@ static int i40e_set_channels(struct net_device *dev,
  	if (vsi->type != I40E_VSI_MAIN)
  		return -EINVAL;
 
+	/* We do not support setting channels via ethtool when TCs are
+	 * configured through mqprio
+	 */
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return -EINVAL;
+
  	/* verify they are not requesting separate vectors */
  	if (!count || ch->rx_count || ch->tx_count)
  		return -EINVAL;
drivers/net/ethernet/intel/i40e/i40e_main.c +69 -0
index 64769cb..5b783fc 100644
@@ -1589,6 +1589,169 @@ static int i40e_set_mac(struct net_device *netdev, void *p)
  }
 
  /**
+ * i40e_config_rss_aq - Prepare for RSS using AQ commands
+ * @vsi: vsi structure
+ * @seed: RSS hash seed
+ **/
+static int i40e_config_rss_aq(struct i40e_vsi *vsi, const u8 *seed,
+			      u8 *lut, u16 lut_size)
+{
+	struct i40e_pf *pf = vsi->back;
+	struct i40e_hw *hw = &pf->hw;
+	int ret = 0;
+
+	if (seed) {
+		struct i40e_aqc_get_set_rss_key_data *seed_dw =
+			(struct i40e_aqc_get_set_rss_key_data *)seed;
+		ret = i40e_aq_set_rss_key(hw, vsi->id, seed_dw);
+		if (ret) {
+			dev_info(&pf->pdev->dev,
+				 "Cannot set RSS key, err %s aq_err %s\n",
+				 i40e_stat_str(hw, ret),
+				 i40e_aq_str(hw, hw->aq.asq_last_status));
+			return ret;
+		}
+	}
+	if (lut) {
+		bool pf_lut = vsi->type == I40E_VSI_MAIN ? true : false;
+
+		ret = i40e_aq_set_rss_lut(hw, vsi->id, pf_lut, lut, lut_size);
+		if (ret) {
+			dev_info(&pf->pdev->dev,
+				 "Cannot set RSS lut, err %s aq_err %s\n",
+				 i40e_stat_str(hw, ret),
+				 i40e_aq_str(hw, hw->aq.asq_last_status));
+			return ret;
+		}
+	}
+	return ret;
+}
+
+/**
+ * i40e_vsi_config_rss - Prepare for VSI(VMDq) RSS if used
+ * @vsi: VSI structure
+ **/
+static int i40e_vsi_config_rss(struct i40e_vsi *vsi)
+{
+	u8 seed[I40E_HKEY_ARRAY_SIZE];
+	struct i40e_pf *pf = vsi->back;
+	u8 *lut;
+	int ret;
+
+	if (!(pf->hw_features & I40E_HW_RSS_AQ_CAPABLE))
+		return 0;
+	if (!vsi->rss_size)
+		vsi->rss_size = min_t(int, pf->alloc_rss_size,
+				      vsi->num_queue_pairs);
+	if (!vsi->rss_size)
+		return -EINVAL;
+	lut = kzalloc(vsi->rss_table_size, GFP_KERNEL);
+	if (!lut)
+		return -ENOMEM;
+
+	/* Use the user configured hash keys and lookup table if there is one,
+	 * otherwise use default
+	 */
+	if (vsi->rss_lut_user)
+		memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size);
+	else
+		i40e_fill_rss_lut(pf, lut, vsi->rss_table_size, vsi->rss_size);
+	if (vsi->rss_hkey_user)
+		memcpy(seed, vsi->rss_hkey_user, I40E_HKEY_ARRAY_SIZE);

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_set_mac(struct net_device *netdev, void *p) · line 1670

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+	else
+		netdev_rss_key_fill((void *)seed, I40E_HKEY_ARRAY_SIZE);
+	ret = i40e_config_rss_aq(vsi, seed, lut, vsi->rss_table_size);
+	kfree(lut);
+	return ret;
+}
+
+/**
+ * i40e_vsi_setup_queue_map_mqprio - Prepares VSI tc_config to have queue
+ * configurations based on MQPRIO options.

This should only be a one line description

Quoted 366 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +254 -23
+ * @vsi: the VSI being configured,
+ * @ctxt: VSI context structure
+ * @enabled_tc: number of traffic classes to enable
+ **/
+static int i40e_vsi_setup_queue_map_mqprio(struct i40e_vsi *vsi,
+					   struct i40e_vsi_context *ctxt,
+					   u8 enabled_tc)
+{
+	u16 qcount = 0, max_qcount, qmap, sections = 0;
+	int i, override_q, pow, num_qps, ret;
+	u8 netdev_tc = 0, offset = 0;
+
+	if (vsi->type != I40E_VSI_MAIN)
+		return -EINVAL;
+	sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID;
+	sections |= I40E_AQ_VSI_PROP_SCHED_VALID;
+	vsi->tc_config.numtc = vsi->mqprio_qopt.qopt.num_tc;
+	vsi->tc_config.enabled_tc = enabled_tc ? enabled_tc : 1;
+	num_qps = vsi->mqprio_qopt.qopt.count[0];
+
+	/* find the next higher power-of-2 of num queue pairs */
+	pow = ilog2(num_qps);
+	if (!is_power_of_2(num_qps))
+		pow++;
+	qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) |
+		(pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT);
+
+	/* Setup queue offset/count for all TCs for given VSI */
+	max_qcount = vsi->mqprio_qopt.qopt.count[0];
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+		/* See if the given TC is enabled for the given VSI */
+		if (vsi->tc_config.enabled_tc & BIT(i)) {
+			offset = vsi->mqprio_qopt.qopt.offset[i];
+			qcount = vsi->mqprio_qopt.qopt.count[i];
+			if (qcount > max_qcount)
+				max_qcount = qcount;
+			vsi->tc_config.tc_info[i].qoffset = offset;
+			vsi->tc_config.tc_info[i].qcount = qcount;
+			vsi->tc_config.tc_info[i].netdev_tc = netdev_tc++;
+		} else {
+			/* TC is not enabled so set the offset to
+			 * default queue and allocate one queue
+			 * for the given TC.
+			 */
+			vsi->tc_config.tc_info[i].qoffset = 0;
+			vsi->tc_config.tc_info[i].qcount = 1;
+			vsi->tc_config.tc_info[i].netdev_tc = 0;
+		}
+	}
+
+	/* Set actual Tx/Rx queue pairs */
+	vsi->num_queue_pairs = offset + qcount;
+
+	/* Setup queue TC[0].qmap for given VSI context */
+	ctxt->info.tc_mapping[0] = cpu_to_le16(qmap);
+	ctxt->info.mapping_flags |= cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG);
+	ctxt->info.queue_mapping[0] = cpu_to_le16(vsi->base_queue);
+	ctxt->info.valid_sections |= cpu_to_le16(sections);
+
+	/* Reconfigure RSS for main VSI with max queue count */
+	vsi->rss_size = max_qcount;
+	ret = i40e_vsi_config_rss(vsi);
+	if (ret) {
+		dev_info(&vsi->back->pdev->dev,
+			 "Failed to reconfig rss for num_queues (%u)\n",
+			 max_qcount);
+		return ret;
+	}
+	vsi->reconfig_rss = true;
+	dev_dbg(&vsi->back->pdev->dev,
+		"Reconfigured rss with num_queues (%u)\n", max_qcount);
+
+	/* Find queue count available for channel VSIs and starting offset
+	 * for channel VSIs
+	 */
+	override_q = vsi->mqprio_qopt.qopt.count[0];
+	if (override_q && (override_q < vsi->num_queue_pairs)) {
+		vsi->cnt_q_avail = vsi->num_queue_pairs - override_q;
+		atomic_set(&vsi->next_base_queue, override_q);
+	}
+	return 0;
+}
+
+/**
   * i40e_vsi_setup_queue_map - Setup a VSI queue map based on enabled_tc
   * @vsi: the VSI being setup
   * @ctxt: VSI context structure
@@ -1626,7 +1789,7 @@ static void i40e_vsi_setup_queue_map(struct i40e_vsi *vsi,
  			numtc = 1;
  		}
  	} else {
-		/* At least TC0 is enabled in case of non-DCB case */
+		/* At least TC0 is enabled in non-DCB, non-MQPRIO case */
  		numtc = 1;
  	}
 
@@ -3168,6 +3331,7 @@ static void i40e_vsi_config_dcb_rings(struct i40e_vsi *vsi)
  			rx_ring->dcb_tc = 0;
  			tx_ring->dcb_tc = 0;
  		}
+		return;
  	}
 
  	for (n = 0; n < I40E_MAX_TRAFFIC_CLASS; n++) {
@@ -4880,6 +5044,24 @@ static u8 i40e_dcb_get_enabled_tc(struct i40e_dcbx_config *dcbcfg)
  }
 
  /**
+ * i40e_mqprio_get_enabled_tc - Get enabled traffic classes
+ * @pf: PF being queried
+ *
+ * Query the current MQPRIO configuration and return the number of
+ * traffic classes enabled.
+ **/
+static u8 i40e_mqprio_get_enabled_tc(struct i40e_pf *pf)
+{
+	struct i40e_vsi *vsi = pf->vsi[pf->lan_vsi];
+	u8 num_tc = vsi->mqprio_qopt.qopt.num_tc;
+	u8 enabled_tc = 1, i;
+
+	for (i = 1; i < num_tc; i++)
+		enabled_tc |= BIT(i);
+	return enabled_tc;
+}
+
+/**
   * i40e_pf_get_num_tc - Get enabled traffic classes for PF
   * @pf: PF being queried
   *
@@ -4892,7 +5074,10 @@ static u8 i40e_pf_get_num_tc(struct i40e_pf *pf)
  	u8 num_tc = 0;
  	struct i40e_dcbx_config *dcbcfg = &hw->local_dcbx_config;
 
-	/* If DCB is not enabled then always in single TC */
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return pf->vsi[pf->lan_vsi]->mqprio_qopt.qopt.num_tc;
+
+	/* If neither MQPRIO nor DCB is enabled, then always use single TC */
  	if (!(pf->flags & I40E_FLAG_DCB_ENABLED))
  		return 1;
 
@@ -4921,7 +5106,12 @@ static u8 i40e_pf_get_num_tc(struct i40e_pf *pf)
   **/
  static u8 i40e_pf_get_tc_map(struct i40e_pf *pf)
  {
-	/* If DCB is not enabled for this PF then just return default TC */
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return i40e_mqprio_get_enabled_tc(pf);
+
+	/* If neither MQPRIO nor DCB is enabled for this PF then just return
+	 * default TC
+	 */
  	if (!(pf->flags & I40E_FLAG_DCB_ENABLED))
  		return I40E_DEFAULT_TRAFFIC_CLASS;
 
@@ -5011,6 +5201,9 @@ static int i40e_vsi_configure_bw_alloc(struct i40e_vsi *vsi, u8 enabled_tc,
  	i40e_status ret;
  	int i;
 
+	if ((vsi->back->flags & I40E_FLAG_TC_MQPRIO) ||
+	    !vsi->mqprio_qopt.qopt.hw)
+		return 0;
  	bw_data.tc_valid_bits = enabled_tc;
  	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
  		bw_data.tc_bw_credits[i] = bw_share[i];
@@ -5073,6 +5266,9 @@ static void i40e_vsi_config_netdev_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  					vsi->tc_config.tc_info[i].qoffset);
  	}
 
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return;
+
  	/* Assign UP2TC map for the VSI */
  	for (i = 0; i < I40E_MAX_USER_PRIORITY; i++) {
  		/* Get the actual TC# for the UP */
@@ -5123,7 +5319,8 @@ static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  	int i;
 
  	/* Check if enabled_tc is same as existing or new TCs */
-	if (vsi->tc_config.enabled_tc == enabled_tc)
+	if (vsi->tc_config.enabled_tc == enabled_tc &&
+	    vsi->mqprio_qopt.qopt.hw != TC_MQPRIO_HW_OFFLOAD)
  		return ret;
 
  	/* Enable ETS TCs with equal BW Share for now across all VSIs */
@@ -5146,15 +5343,37 @@ static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  	ctxt.vf_num = 0;
  	ctxt.uplink_seid = vsi->uplink_seid;
  	ctxt.info = vsi->info;
-	i40e_vsi_setup_queue_map(vsi, &ctxt, enabled_tc, false);
+	if (vsi->back->flags & I40E_FLAG_TC_MQPRIO) {
+		ret = i40e_vsi_setup_queue_map_mqprio(vsi, &ctxt, enabled_tc);
+		if (ret)
+			goto out;
+	} else {
+		i40e_vsi_setup_queue_map(vsi, &ctxt, enabled_tc, false);
+	}
 
+	/* On destroying the qdisc, reset vsi->rss_size, as number of enabled
+	 * queues changed.
+	 */
+	if (!vsi->mqprio_qopt.qopt.hw && vsi->reconfig_rss) {
+		vsi->rss_size = min_t(int, vsi->back->alloc_rss_size,
+				      vsi->num_queue_pairs);
+		ret = i40e_vsi_config_rss(vsi);
+		if (ret) {
+			dev_info(&vsi->back->pdev->dev,
+				 "Failed to reconfig rss for num_queues\n");
+			return ret;
+		}
+		vsi->reconfig_rss = false;
+	}
  	if (vsi->back->flags & I40E_FLAG_IWARP_ENABLED) {
  		ctxt.info.valid_sections |=
  				cpu_to_le16(I40E_AQ_VSI_PROP_QUEUE_OPT_VALID);
  		ctxt.info.queueing_opt_flags |= I40E_AQ_VSI_QUE_OPT_TCP_ENA;
  	}
 
-	/* Update the VSI after updating the VSI queue-mapping information */
+	/* Update the VSI after updating the VSI queue-mapping
+	 * information
+	 */
  	ret = i40e_aq_update_vsi_params(&vsi->back->hw, &ctxt, NULL);
  	if (ret) {
  		dev_info(&vsi->back->pdev->dev,
@@ -6252,54 +6471,170 @@ void i40e_down(struct i40e_vsi *vsi)
  }
 
  /**
+ * i40e_validate_mqprio_qopt- validate queue mapping info
+ * @vsi: the VSI being configured
+ * @mqprio_qopt: queue parametrs
+ **/
+static int i40e_validate_mqprio_qopt(struct i40e_vsi *vsi,
+				     struct tc_mqprio_qopt_offload *mqprio_qopt)
+{
+	int i;
+
+	if ((mqprio_qopt->qopt.offset[0] != 0) ||
+	    (mqprio_qopt->qopt.num_tc < 1) ||
+	    (mqprio_qopt->qopt.num_tc > I40E_MAX_TRAFFIC_CLASS))
+		return -EINVAL;
+	for (i = 0; ; i++) {
+		if (!mqprio_qopt->qopt.count[i])
+			return -EINVAL;
+		if (mqprio_qopt->min_rate[i] || mqprio_qopt->max_rate[i])
+			return -EINVAL;
+		if (i >= mqprio_qopt->qopt.num_tc - 1)
+			break;
+		if (mqprio_qopt->qopt.offset[i + 1] !=
+		    (mqprio_qopt->qopt.offset[i] + mqprio_qopt->qopt.count[i]))
+			return -EINVAL;
+	}
+	if (vsi->num_queue_pairs <
+	    (mqprio_qopt->qopt.offset[i] + mqprio_qopt->qopt.count[i])) {
+		return -EINVAL;
+	}
+	return 0;
+}
+
+/**
+ * i40e_vsi_set_default_tc_config - set default values for tc configuration
+ * @vsi: the VSI being configured
+ **/
+static void i40e_vsi_set_default_tc_config(struct i40e_vsi *vsi)
+{
+	u16 qcount;
+	int i;
+
+	/* Only TC0 is enabled */
+	vsi->tc_config.numtc = 1;
+	vsi->tc_config.enabled_tc = 1;
+	qcount = min_t(int, vsi->alloc_queue_pairs,
+		       i40e_pf_get_max_q_per_tc(vsi->back));
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+		/* For the TC that is not enabled set the offset to to default
+		 * queue and allocate one queue for the given TC.
+		 */
+		vsi->tc_config.tc_info[i].qoffset = 0;
+		if (i == 0)
+			vsi->tc_config.tc_info[i].qcount = qcount;
+		else
+			vsi->tc_config.tc_info[i].qcount = 1;
+		vsi->tc_config.tc_info[i].netdev_tc = 0;
+	}
+}
+
+/**
   * i40e_setup_tc - configure multiple traffic classes
   * @netdev: net device to configure
- * @tc: number of traffic classes to enable
+ * @type: type of tc offload
+ * @type_data: offload data
   **/
-static int i40e_setup_tc(struct net_device *netdev, u8 tc)
+static int i40e_setup_tc(struct net_device *netdev, enum tc_setup_type type,
+			 void *type_data)
  {
  	struct i40e_netdev_priv *np = netdev_priv(netdev);
+	struct tc_mqprio_qopt_offload *mqprio_qopt = NULL;
  	struct i40e_vsi *vsi = np->vsi;
  	struct i40e_pf *pf = vsi->back;
-	u8 enabled_tc = 0;
+	struct tc_mqprio_qopt *mqprio = NULL;
+	u8 enabled_tc = 0, num_tc = 0, hw = 0;
  	int ret = -EINVAL;
  	int i;
 
-	/* Check if DCB enabled to continue */
-	if (!(pf->flags & I40E_FLAG_DCB_ENABLED)) {
-		netdev_info(netdev, "DCB is not enabled for adapter\n");
-		goto exit;
-	}
+	if (type == TC_SETUP_MQPRIO) {
+		mqprio = type_data;
+		hw = mqprio->hw;
+		num_tc = mqprio->num_tc;
+	} else if (type == TC_SETUP_MQPRIO_EXT) {
+		mqprio_qopt = type_data;
+		hw = mqprio_qopt->qopt.hw;
+		num_tc = mqprio_qopt->qopt.num_tc;
+	}
+	if (!hw) {
+		pf->flags &= ~I40E_FLAG_TC_MQPRIO;
+		if (type == TC_SETUP_MQPRIO_EXT)
+			memcpy(&vsi->mqprio_qopt, mqprio_qopt,
+			       sizeof(*mqprio_qopt));
+		goto config_tc;
+	}
+	switch (hw) {
+	case TC_MQPRIO_HW_OFFLOAD_TCS:
+		pf->flags &= ~I40E_FLAG_TC_MQPRIO;
+
+		/* Check if DCB enabled to continue */
+		if (!(pf->flags & I40E_FLAG_DCB_ENABLED)) {
+			netdev_info(netdev,
+				    "DCB is not enabled for adapter\n");
+			goto exit;
+		}
 
-	/* Check if MFP enabled */
-	if (pf->flags & I40E_FLAG_MFP_ENABLED) {
-		netdev_info(netdev, "Configuring TC not supported in MFP mode\n");
-		goto exit;
-	}
+		/* Check if MFP enabled */
+		if (pf->flags & I40E_FLAG_MFP_ENABLED) {
+			netdev_info(netdev,
+				    "Configuring TC not supported in MFP mode\n");
+			goto exit;
+		}
 
-	/* Check whether tc count is within enabled limit */
-	if (tc > i40e_pf_get_num_tc(pf)) {
-		netdev_info(netdev, "TC count greater than enabled on link for adapter\n");
-		goto exit;
+		/* Check whether tc count is within enabled limit */
+		if (num_tc > i40e_pf_get_num_tc(pf)) {
+			netdev_info(netdev,
+				    "TC count greater than enabled on link for adapter\n");
+			goto exit;
+		}
+		break;
+	case TC_MQPRIO_HW_OFFLOAD:
+		if (pf->flags & I40E_FLAG_DCB_ENABLED) {
+			netdev_info(netdev,

drivers/net/ethernet/intel/i40e/i40e_main.c void i40e_down(struct i40e_vsi *vsi) · line 6603

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+				    "Full offload of TC Mqprio options is not supported when DCB is enabled\n");
+			goto exit;
+		}
+
+		/* Check if MFP enabled */
+		if (pf->flags & I40E_FLAG_MFP_ENABLED) {
+			netdev_info(netdev,
+				    "Configuring TC not supported in MFP mode\n");
+			goto exit;
+		}

This check is in both cases, and probably always needs to be checked, so it can probably be generalized and put once before this switch rather than repeating the code.

Quoted 41 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +18 -3
+		ret = i40e_validate_mqprio_qopt(vsi, mqprio_qopt);
+		if (ret)
+			goto exit;
+		memcpy(&vsi->mqprio_qopt, mqprio_qopt,
+		       sizeof(*mqprio_qopt));
+		pf->flags |= I40E_FLAG_TC_MQPRIO;
+		pf->flags &= ~I40E_FLAG_DCB_ENABLED;
+		break;
+	default:
+		return -EINVAL;
  	}
 
+config_tc:
  	/* Generate TC map for number of tc requested */
-	for (i = 0; i < tc; i++)
+	for (i = 0; i < num_tc; i++)
  		enabled_tc |= BIT(i);
 
  	/* Requesting same TC configuration as already enabled */
-	if (enabled_tc == vsi->tc_config.enabled_tc)
+	if (enabled_tc == vsi->tc_config.enabled_tc &&
+	    hw != TC_MQPRIO_HW_OFFLOAD)
  		return 0;
 
  	/* Quiesce VSI queues */
  	i40e_quiesce_vsi(vsi);
 
+	if (!hw && !(pf->flags & I40E_FLAG_TC_MQPRIO))
+		i40e_remove_queue_channels(vsi);
+
  	/* Configure VSI for enabled TCs */
  	ret = i40e_vsi_config_tc(vsi, enabled_tc);
  	if (ret) {
  		netdev_info(netdev, "Failed configuring TC for VSI seid=%d\n",
  			    vsi->seid);
-		goto exit;
+		goto reset;
  	}
 
  	if (pf->flags & I40E_FLAG_TC_MQPRIO) {

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_setup_tc(struct net_device *netdev, u8 tc) · line 6649

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +2 -2
@@ -6307,27 +6642,31 @@ static int i40e_setup_tc(struct net_device *netdev, u8 tc)
  		if (ret) {
  			netdev_info(netdev,
  				    "Failed configuring queue channels\n");
-			goto exit;
+			goto reset;
  		}
  	}
 
-exit:
+	goto out;

This "goto out" just to skip the reset call seems rather ugly. There's got to be a better way to do this and clean up the spaghetti, for the error cases and for the normal case.

If there is no need to do cleanup, all the "goto exit" calls could be "return ret", which gets rid of the "exit" label.

Maybe the need for unquiesce can be tracked by a bool?

Quoted 129 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +11 -86
+reset:
+	/* Reset the configuration data to defaults, only TC0 is enabled */
+	i40e_vsi_set_default_tc_config(vsi);
+out:
  	/* Unquiesce VSI */
  	i40e_unquiesce_vsi(vsi);
+exit:
  	return ret;
  }
 
  static int __i40e_setup_tc(struct net_device *netdev, enum tc_setup_type type,
  			   void *type_data)
  {
-	struct tc_mqprio_qopt *mqprio = type_data;
-
-	if (type != TC_SETUP_MQPRIO)
+	switch (type) {
+	case TC_SETUP_MQPRIO:
+	case TC_SETUP_MQPRIO_EXT:
+		return i40e_setup_tc(netdev, type, type_data);
+	default:
  		return -EOPNOTSUPP;
-
-	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
-
-	return i40e_setup_tc(netdev, mqprio->num_tc);
+	}
  }
 
  /**
@@ -9174,45 +9513,6 @@ static int i40e_setup_misc_vector(struct i40e_pf *pf)
  }
 
  /**
- * i40e_config_rss_aq - Prepare for RSS using AQ commands
- * @vsi: vsi structure
- * @seed: RSS hash seed
- **/
-static int i40e_config_rss_aq(struct i40e_vsi *vsi, const u8 *seed,
-			      u8 *lut, u16 lut_size)
-{
-	struct i40e_pf *pf = vsi->back;
-	struct i40e_hw *hw = &pf->hw;
-	int ret = 0;
-
-	if (seed) {
-		struct i40e_aqc_get_set_rss_key_data *seed_dw =
-			(struct i40e_aqc_get_set_rss_key_data *)seed;
-		ret = i40e_aq_set_rss_key(hw, vsi->id, seed_dw);
-		if (ret) {
-			dev_info(&pf->pdev->dev,
-				 "Cannot set RSS key, err %s aq_err %s\n",
-				 i40e_stat_str(hw, ret),
-				 i40e_aq_str(hw, hw->aq.asq_last_status));
-			return ret;
-		}
-	}
-	if (lut) {
-		bool pf_lut = vsi->type == I40E_VSI_MAIN ? true : false;
-
-		ret = i40e_aq_set_rss_lut(hw, vsi->id, pf_lut, lut, lut_size);
-		if (ret) {
-			dev_info(&pf->pdev->dev,
-				 "Cannot set RSS lut, err %s aq_err %s\n",
-				 i40e_stat_str(hw, ret),
-				 i40e_aq_str(hw, hw->aq.asq_last_status));
-			return ret;
-		}
-	}
-	return ret;
-}
-
-/**
   * i40e_get_rss_aq - Get RSS keys and lut by using AQ commands
   * @vsi: Pointer to vsi structure
   * @seed: Buffter to store the hash keys
@@ -9259,46 +9559,6 @@ static int i40e_get_rss_aq(struct i40e_vsi *vsi, const u8 *seed,
  }
 
  /**
- * i40e_vsi_config_rss - Prepare for VSI(VMDq) RSS if used
- * @vsi: VSI structure
- **/
-static int i40e_vsi_config_rss(struct i40e_vsi *vsi)
-{
-	u8 seed[I40E_HKEY_ARRAY_SIZE];
-	struct i40e_pf *pf = vsi->back;
-	u8 *lut;
-	int ret;
-
-	if (!(pf->hw_features & I40E_HW_RSS_AQ_CAPABLE))
-		return 0;
-
-	if (!vsi->rss_size)
-		vsi->rss_size = min_t(int, pf->alloc_rss_size,
-				      vsi->num_queue_pairs);
-	if (!vsi->rss_size)
-		return -EINVAL;
-
-	lut = kzalloc(vsi->rss_table_size, GFP_KERNEL);
-	if (!lut)
-		return -ENOMEM;
-	/* Use the user configured hash keys and lookup table if there is one,
-	 * otherwise use default
-	 */
-	if (vsi->rss_lut_user)
-		memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size);
-	else
-		i40e_fill_rss_lut(pf, lut, vsi->rss_table_size, vsi->rss_size);
-	if (vsi->rss_hkey_user)
-		memcpy(seed, vsi->rss_hkey_user, I40E_HKEY_ARRAY_SIZE);
-	else
-		netdev_rss_key_fill((void *)seed, I40E_HKEY_ARRAY_SIZE);
-	ret = i40e_config_rss_aq(vsi, seed, lut, vsi->rss_table_size);
-	kfree(lut);
-
-	return ret;
-}
-
-/**
   * i40e_config_rss_reg - Configure RSS keys and lut by writing registers
   * @vsi: Pointer to vsi structure
   * @seed: RSS hash seed

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

#10 lore raw

PATCH v2 4/6 net: i40e: Enable mqprio full offload mode in the i40e driver for configuring TCs and queue mapping

Quoted 187 lines · Shannon Nelson
Quoted 186 lines · Amritha Nambiar

The i40e driver is modified to enable the new mqprio hardware offload mode and factor the TCs and queue configuration by creating channel VSIs. In this mode, the priority to traffic class mapping and the user specified queue ranges are used to configure the traffic classes when the 'hw' option is set to 2.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 4\
   map 0 0 0 0 1 2 2 3 queues 2 at 0 2 at 2 1 at 4 1 at 5 hw 2
# tc qdisc show dev eth0

qdisc mqprio 8038: root tc 4 map 0 0 0 0 1 2 2 3 0 0 0 0 0 0 0 0

              queues:(0:1) (2:3) (4:4) (5:5)

The HW channels created are removed and all the queue configuration is set to default when the qdisc is detached from the root of the device.

# tc qdisc del dev eth0 root

This patch also disables setting up channels via ethtool (ethtool -L) when the TCs are configured using mqprio scheduler.

The patch also limits setting ethtool Rx flow hash indirection (ethtool -X eth0 equal N) to max queues configured via mqprio. The Rx flow hash indirection input through ethtool should be validated so that it is within in the queue range configured via tc/mqprio. The bound checking is achieved by reporting the current rss size to the kernel when queues are configured via mqprio.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 4\
   map 0 0 0 1 0 2 3 0 queues 2 at 0 4 at 2 8 at 6 11 at 14 hw 2
# ethtool -X eth0 equal 12
Cannot set RX flow hash configuration: Invalid argument

v2: Clean up __i40e_setup_tc() and i40e_setup_tc() to work with Jiri's changes to the ndo_setup_tc interface which now takes a type and the type_data for the offload. No need to disable ATR in MQPRIO mode.

---
  drivers/net/ethernet/intel/i40e/i40e.h         |    3
  drivers/net/ethernet/intel/i40e/i40e_ethtool.c |    8
  drivers/net/ethernet/intel/i40e/i40e_main.c    |  488 ++++++++++++++++++------
  3 files changed, 384 insertions(+), 115 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +3 -0
index e259a67..2565280 100644
@@ -54,6 +54,7 @@
  #include <linux/clocksource.h>
  #include <linux/net_tstamp.h>
  #include <linux/ptp_clock_kernel.h>
+#include <net/pkt_cls.h>
  #include "i40e_type.h"
  #include "i40e_prototype.h"
  #include "i40e_client.h"
@@ -699,6 +700,7 @@ struct i40e_vsi {
  	enum i40e_vsi_type type;  /* VSI type, e.g., LAN, FCoE, etc */
  	s16 vf_id;		/* Virtual function ID for SRIOV VSIs */
 
+	struct tc_mqprio_qopt_offload mqprio_qopt; /* queue parameters */
  	struct i40e_tc_configuration tc_config;
  	struct i40e_aqc_vsi_properties_data info;
 
@@ -724,6 +726,7 @@ struct i40e_vsi {
  	u16 cnt_q_avail; /* num of queues available for channel usage */
  	u16 orig_rss_size;
  	u16 current_rss_size;
+	bool reconfig_rss;
 
  	/* keeps track of next_base_queue to be used for channel setup */
  	atomic_t next_base_queue;
drivers/net/ethernet/intel/i40e/i40e_ethtool.c +7 -1
index b273c24..326fc18 100644
@@ -2647,7 +2647,7 @@ static int i40e_get_rxnfc(struct net_device *netdev, struct ethtool_rxnfc *cmd,
 
  	switch (cmd->cmd) {
  	case ETHTOOL_GRXRINGS:
-		cmd->data = vsi->num_queue_pairs;
+		cmd->data = vsi->rss_size;
  		ret = 0;
  		break;
  	case ETHTOOL_GRXFH:
@@ -3892,6 +3892,12 @@ static int i40e_set_channels(struct net_device *dev,
  	if (vsi->type != I40E_VSI_MAIN)
  		return -EINVAL;
 
+	/* We do not support setting channels via ethtool when TCs are
+	 * configured through mqprio
+	 */
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return -EINVAL;
+
  	/* verify they are not requesting separate vectors */
  	if (!count || ch->rx_count || ch->tx_count)
  		return -EINVAL;
drivers/net/ethernet/intel/i40e/i40e_main.c +73 -0
index 64769cb..5b783fc 100644
@@ -1589,6 +1589,169 @@ static int i40e_set_mac(struct net_device *netdev, void *p)
  }
 
  /**
+ * i40e_config_rss_aq - Prepare for RSS using AQ commands
+ * @vsi: vsi structure
+ * @seed: RSS hash seed
+ **/
+static int i40e_config_rss_aq(struct i40e_vsi *vsi, const u8 *seed,
+			      u8 *lut, u16 lut_size)
+{
+	struct i40e_pf *pf = vsi->back;
+	struct i40e_hw *hw = &pf->hw;
+	int ret = 0;
+
+	if (seed) {
+		struct i40e_aqc_get_set_rss_key_data *seed_dw =
+			(struct i40e_aqc_get_set_rss_key_data *)seed;
+		ret = i40e_aq_set_rss_key(hw, vsi->id, seed_dw);
+		if (ret) {
+			dev_info(&pf->pdev->dev,
+				 "Cannot set RSS key, err %s aq_err %s\n",
+				 i40e_stat_str(hw, ret),
+				 i40e_aq_str(hw, hw->aq.asq_last_status));
+			return ret;
+		}
+	}
+	if (lut) {
+		bool pf_lut = vsi->type == I40E_VSI_MAIN ? true : false;
+
+		ret = i40e_aq_set_rss_lut(hw, vsi->id, pf_lut, lut, lut_size);
+		if (ret) {
+			dev_info(&pf->pdev->dev,
+				 "Cannot set RSS lut, err %s aq_err %s\n",
+				 i40e_stat_str(hw, ret),
+				 i40e_aq_str(hw, hw->aq.asq_last_status));
+			return ret;
+		}
+	}
+	return ret;
+}
+
+/**
+ * i40e_vsi_config_rss - Prepare for VSI(VMDq) RSS if used
+ * @vsi: VSI structure
+ **/
+static int i40e_vsi_config_rss(struct i40e_vsi *vsi)
+{
+	u8 seed[I40E_HKEY_ARRAY_SIZE];
+	struct i40e_pf *pf = vsi->back;
+	u8 *lut;
+	int ret;
+
+	if (!(pf->hw_features & I40E_HW_RSS_AQ_CAPABLE))
+		return 0;
+	if (!vsi->rss_size)
+		vsi->rss_size = min_t(int, pf->alloc_rss_size,
+				      vsi->num_queue_pairs);
+	if (!vsi->rss_size)
+		return -EINVAL;
+	lut = kzalloc(vsi->rss_table_size, GFP_KERNEL);
+	if (!lut)
+		return -ENOMEM;
+
+	/* Use the user configured hash keys and lookup table if there is one,
+	 * otherwise use default
+	 */
+	if (vsi->rss_lut_user)
+		memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size);
+	else
+		i40e_fill_rss_lut(pf, lut, vsi->rss_table_size, vsi->rss_size);
+	if (vsi->rss_hkey_user)
+		memcpy(seed, vsi->rss_hkey_user, I40E_HKEY_ARRAY_SIZE);
+	else
+		netdev_rss_key_fill((void *)seed, I40E_HKEY_ARRAY_SIZE);
+	ret = i40e_config_rss_aq(vsi, seed, lut, vsi->rss_table_size);
+	kfree(lut);
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_set_mac(struct net_device *netdev, void *p) · line 1670

Quoted 6 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +6 -0
+	return ret;
+}
+
+/**
+ * i40e_vsi_setup_queue_map_mqprio - Prepares VSI tc_config to have queue
+ * configurations based on MQPRIO options.

This should only be a one line description

Will fix in v3.

Quoted 373 lines
Quoted 373 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +260 -23
+ * @vsi: the VSI being configured,
+ * @ctxt: VSI context structure
+ * @enabled_tc: number of traffic classes to enable
+ **/
+static int i40e_vsi_setup_queue_map_mqprio(struct i40e_vsi *vsi,
+					   struct i40e_vsi_context *ctxt,
+					   u8 enabled_tc)
+{
+	u16 qcount = 0, max_qcount, qmap, sections = 0;
+	int i, override_q, pow, num_qps, ret;
+	u8 netdev_tc = 0, offset = 0;
+
+	if (vsi->type != I40E_VSI_MAIN)
+		return -EINVAL;
+	sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID;
+	sections |= I40E_AQ_VSI_PROP_SCHED_VALID;
+	vsi->tc_config.numtc = vsi->mqprio_qopt.qopt.num_tc;
+	vsi->tc_config.enabled_tc = enabled_tc ? enabled_tc : 1;
+	num_qps = vsi->mqprio_qopt.qopt.count[0];
+
+	/* find the next higher power-of-2 of num queue pairs */
+	pow = ilog2(num_qps);
+	if (!is_power_of_2(num_qps))
+		pow++;
+	qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) |
+		(pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT);
+
+	/* Setup queue offset/count for all TCs for given VSI */
+	max_qcount = vsi->mqprio_qopt.qopt.count[0];
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+		/* See if the given TC is enabled for the given VSI */
+		if (vsi->tc_config.enabled_tc & BIT(i)) {
+			offset = vsi->mqprio_qopt.qopt.offset[i];
+			qcount = vsi->mqprio_qopt.qopt.count[i];
+			if (qcount > max_qcount)
+				max_qcount = qcount;
+			vsi->tc_config.tc_info[i].qoffset = offset;
+			vsi->tc_config.tc_info[i].qcount = qcount;
+			vsi->tc_config.tc_info[i].netdev_tc = netdev_tc++;
+		} else {
+			/* TC is not enabled so set the offset to
+			 * default queue and allocate one queue
+			 * for the given TC.
+			 */
+			vsi->tc_config.tc_info[i].qoffset = 0;
+			vsi->tc_config.tc_info[i].qcount = 1;
+			vsi->tc_config.tc_info[i].netdev_tc = 0;
+		}
+	}
+
+	/* Set actual Tx/Rx queue pairs */
+	vsi->num_queue_pairs = offset + qcount;
+
+	/* Setup queue TC[0].qmap for given VSI context */
+	ctxt->info.tc_mapping[0] = cpu_to_le16(qmap);
+	ctxt->info.mapping_flags |= cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG);
+	ctxt->info.queue_mapping[0] = cpu_to_le16(vsi->base_queue);
+	ctxt->info.valid_sections |= cpu_to_le16(sections);
+
+	/* Reconfigure RSS for main VSI with max queue count */
+	vsi->rss_size = max_qcount;
+	ret = i40e_vsi_config_rss(vsi);
+	if (ret) {
+		dev_info(&vsi->back->pdev->dev,
+			 "Failed to reconfig rss for num_queues (%u)\n",
+			 max_qcount);
+		return ret;
+	}
+	vsi->reconfig_rss = true;
+	dev_dbg(&vsi->back->pdev->dev,
+		"Reconfigured rss with num_queues (%u)\n", max_qcount);
+
+	/* Find queue count available for channel VSIs and starting offset
+	 * for channel VSIs
+	 */
+	override_q = vsi->mqprio_qopt.qopt.count[0];
+	if (override_q && (override_q < vsi->num_queue_pairs)) {
+		vsi->cnt_q_avail = vsi->num_queue_pairs - override_q;
+		atomic_set(&vsi->next_base_queue, override_q);
+	}
+	return 0;
+}
+
+/**
   * i40e_vsi_setup_queue_map - Setup a VSI queue map based on enabled_tc
   * @vsi: the VSI being setup
   * @ctxt: VSI context structure
@@ -1626,7 +1789,7 @@ static void i40e_vsi_setup_queue_map(struct i40e_vsi *vsi,
  			numtc = 1;
  		}
  	} else {
-		/* At least TC0 is enabled in case of non-DCB case */
+		/* At least TC0 is enabled in non-DCB, non-MQPRIO case */
  		numtc = 1;
  	}
 
@@ -3168,6 +3331,7 @@ static void i40e_vsi_config_dcb_rings(struct i40e_vsi *vsi)
  			rx_ring->dcb_tc = 0;
  			tx_ring->dcb_tc = 0;
  		}
+		return;
  	}
 
  	for (n = 0; n < I40E_MAX_TRAFFIC_CLASS; n++) {
@@ -4880,6 +5044,24 @@ static u8 i40e_dcb_get_enabled_tc(struct i40e_dcbx_config *dcbcfg)
  }
 
  /**
+ * i40e_mqprio_get_enabled_tc - Get enabled traffic classes
+ * @pf: PF being queried
+ *
+ * Query the current MQPRIO configuration and return the number of
+ * traffic classes enabled.
+ **/
+static u8 i40e_mqprio_get_enabled_tc(struct i40e_pf *pf)
+{
+	struct i40e_vsi *vsi = pf->vsi[pf->lan_vsi];
+	u8 num_tc = vsi->mqprio_qopt.qopt.num_tc;
+	u8 enabled_tc = 1, i;
+
+	for (i = 1; i < num_tc; i++)
+		enabled_tc |= BIT(i);
+	return enabled_tc;
+}
+
+/**
   * i40e_pf_get_num_tc - Get enabled traffic classes for PF
   * @pf: PF being queried
   *
@@ -4892,7 +5074,10 @@ static u8 i40e_pf_get_num_tc(struct i40e_pf *pf)
  	u8 num_tc = 0;
  	struct i40e_dcbx_config *dcbcfg = &hw->local_dcbx_config;
 
-	/* If DCB is not enabled then always in single TC */
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return pf->vsi[pf->lan_vsi]->mqprio_qopt.qopt.num_tc;
+
+	/* If neither MQPRIO nor DCB is enabled, then always use single TC */
  	if (!(pf->flags & I40E_FLAG_DCB_ENABLED))
  		return 1;
 
@@ -4921,7 +5106,12 @@ static u8 i40e_pf_get_num_tc(struct i40e_pf *pf)
   **/
  static u8 i40e_pf_get_tc_map(struct i40e_pf *pf)
  {
-	/* If DCB is not enabled for this PF then just return default TC */
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return i40e_mqprio_get_enabled_tc(pf);
+
+	/* If neither MQPRIO nor DCB is enabled for this PF then just return
+	 * default TC
+	 */
  	if (!(pf->flags & I40E_FLAG_DCB_ENABLED))
  		return I40E_DEFAULT_TRAFFIC_CLASS;
 
@@ -5011,6 +5201,9 @@ static int i40e_vsi_configure_bw_alloc(struct i40e_vsi *vsi, u8 enabled_tc,
  	i40e_status ret;
  	int i;
 
+	if ((vsi->back->flags & I40E_FLAG_TC_MQPRIO) ||
+	    !vsi->mqprio_qopt.qopt.hw)
+		return 0;
  	bw_data.tc_valid_bits = enabled_tc;
  	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
  		bw_data.tc_bw_credits[i] = bw_share[i];
@@ -5073,6 +5266,9 @@ static void i40e_vsi_config_netdev_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  					vsi->tc_config.tc_info[i].qoffset);
  	}
 
+	if (pf->flags & I40E_FLAG_TC_MQPRIO)
+		return;
+
  	/* Assign UP2TC map for the VSI */
  	for (i = 0; i < I40E_MAX_USER_PRIORITY; i++) {
  		/* Get the actual TC# for the UP */
@@ -5123,7 +5319,8 @@ static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  	int i;
 
  	/* Check if enabled_tc is same as existing or new TCs */
-	if (vsi->tc_config.enabled_tc == enabled_tc)
+	if (vsi->tc_config.enabled_tc == enabled_tc &&
+	    vsi->mqprio_qopt.qopt.hw != TC_MQPRIO_HW_OFFLOAD)
  		return ret;
 
  	/* Enable ETS TCs with equal BW Share for now across all VSIs */
@@ -5146,15 +5343,37 @@ static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  	ctxt.vf_num = 0;
  	ctxt.uplink_seid = vsi->uplink_seid;
  	ctxt.info = vsi->info;
-	i40e_vsi_setup_queue_map(vsi, &ctxt, enabled_tc, false);
+	if (vsi->back->flags & I40E_FLAG_TC_MQPRIO) {
+		ret = i40e_vsi_setup_queue_map_mqprio(vsi, &ctxt, enabled_tc);
+		if (ret)
+			goto out;
+	} else {
+		i40e_vsi_setup_queue_map(vsi, &ctxt, enabled_tc, false);
+	}
 
+	/* On destroying the qdisc, reset vsi->rss_size, as number of enabled
+	 * queues changed.
+	 */
+	if (!vsi->mqprio_qopt.qopt.hw && vsi->reconfig_rss) {
+		vsi->rss_size = min_t(int, vsi->back->alloc_rss_size,
+				      vsi->num_queue_pairs);
+		ret = i40e_vsi_config_rss(vsi);
+		if (ret) {
+			dev_info(&vsi->back->pdev->dev,
+				 "Failed to reconfig rss for num_queues\n");
+			return ret;
+		}
+		vsi->reconfig_rss = false;
+	}
  	if (vsi->back->flags & I40E_FLAG_IWARP_ENABLED) {
  		ctxt.info.valid_sections |=
  				cpu_to_le16(I40E_AQ_VSI_PROP_QUEUE_OPT_VALID);
  		ctxt.info.queueing_opt_flags |= I40E_AQ_VSI_QUE_OPT_TCP_ENA;
  	}
 
-	/* Update the VSI after updating the VSI queue-mapping information */
+	/* Update the VSI after updating the VSI queue-mapping
+	 * information
+	 */
  	ret = i40e_aq_update_vsi_params(&vsi->back->hw, &ctxt, NULL);
  	if (ret) {
  		dev_info(&vsi->back->pdev->dev,
@@ -6252,54 +6471,170 @@ void i40e_down(struct i40e_vsi *vsi)
  }
 
  /**
+ * i40e_validate_mqprio_qopt- validate queue mapping info
+ * @vsi: the VSI being configured
+ * @mqprio_qopt: queue parametrs
+ **/
+static int i40e_validate_mqprio_qopt(struct i40e_vsi *vsi,
+				     struct tc_mqprio_qopt_offload *mqprio_qopt)
+{
+	int i;
+
+	if ((mqprio_qopt->qopt.offset[0] != 0) ||
+	    (mqprio_qopt->qopt.num_tc < 1) ||
+	    (mqprio_qopt->qopt.num_tc > I40E_MAX_TRAFFIC_CLASS))
+		return -EINVAL;
+	for (i = 0; ; i++) {
+		if (!mqprio_qopt->qopt.count[i])
+			return -EINVAL;
+		if (mqprio_qopt->min_rate[i] || mqprio_qopt->max_rate[i])
+			return -EINVAL;
+		if (i >= mqprio_qopt->qopt.num_tc - 1)
+			break;
+		if (mqprio_qopt->qopt.offset[i + 1] !=
+		    (mqprio_qopt->qopt.offset[i] + mqprio_qopt->qopt.count[i]))
+			return -EINVAL;
+	}
+	if (vsi->num_queue_pairs <
+	    (mqprio_qopt->qopt.offset[i] + mqprio_qopt->qopt.count[i])) {
+		return -EINVAL;
+	}
+	return 0;
+}
+
+/**
+ * i40e_vsi_set_default_tc_config - set default values for tc configuration
+ * @vsi: the VSI being configured
+ **/
+static void i40e_vsi_set_default_tc_config(struct i40e_vsi *vsi)
+{
+	u16 qcount;
+	int i;
+
+	/* Only TC0 is enabled */
+	vsi->tc_config.numtc = 1;
+	vsi->tc_config.enabled_tc = 1;
+	qcount = min_t(int, vsi->alloc_queue_pairs,
+		       i40e_pf_get_max_q_per_tc(vsi->back));
+	for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+		/* For the TC that is not enabled set the offset to to default
+		 * queue and allocate one queue for the given TC.
+		 */
+		vsi->tc_config.tc_info[i].qoffset = 0;
+		if (i == 0)
+			vsi->tc_config.tc_info[i].qcount = qcount;
+		else
+			vsi->tc_config.tc_info[i].qcount = 1;
+		vsi->tc_config.tc_info[i].netdev_tc = 0;
+	}
+}
+
+/**
   * i40e_setup_tc - configure multiple traffic classes
   * @netdev: net device to configure
- * @tc: number of traffic classes to enable
+ * @type: type of tc offload
+ * @type_data: offload data
   **/
-static int i40e_setup_tc(struct net_device *netdev, u8 tc)
+static int i40e_setup_tc(struct net_device *netdev, enum tc_setup_type type,
+			 void *type_data)
  {
  	struct i40e_netdev_priv *np = netdev_priv(netdev);
+	struct tc_mqprio_qopt_offload *mqprio_qopt = NULL;
  	struct i40e_vsi *vsi = np->vsi;
  	struct i40e_pf *pf = vsi->back;
-	u8 enabled_tc = 0;
+	struct tc_mqprio_qopt *mqprio = NULL;
+	u8 enabled_tc = 0, num_tc = 0, hw = 0;
  	int ret = -EINVAL;
  	int i;
 
-	/* Check if DCB enabled to continue */
-	if (!(pf->flags & I40E_FLAG_DCB_ENABLED)) {
-		netdev_info(netdev, "DCB is not enabled for adapter\n");
-		goto exit;
-	}
+	if (type == TC_SETUP_MQPRIO) {
+		mqprio = type_data;
+		hw = mqprio->hw;
+		num_tc = mqprio->num_tc;
+	} else if (type == TC_SETUP_MQPRIO_EXT) {
+		mqprio_qopt = type_data;
+		hw = mqprio_qopt->qopt.hw;
+		num_tc = mqprio_qopt->qopt.num_tc;
+	}
+	if (!hw) {
+		pf->flags &= ~I40E_FLAG_TC_MQPRIO;
+		if (type == TC_SETUP_MQPRIO_EXT)
+			memcpy(&vsi->mqprio_qopt, mqprio_qopt,
+			       sizeof(*mqprio_qopt));
+		goto config_tc;
+	}
+	switch (hw) {
+	case TC_MQPRIO_HW_OFFLOAD_TCS:
+		pf->flags &= ~I40E_FLAG_TC_MQPRIO;
+
+		/* Check if DCB enabled to continue */
+		if (!(pf->flags & I40E_FLAG_DCB_ENABLED)) {
+			netdev_info(netdev,
+				    "DCB is not enabled for adapter\n");
+			goto exit;
+		}
 
-	/* Check if MFP enabled */
-	if (pf->flags & I40E_FLAG_MFP_ENABLED) {
-		netdev_info(netdev, "Configuring TC not supported in MFP mode\n");
-		goto exit;
-	}
+		/* Check if MFP enabled */
+		if (pf->flags & I40E_FLAG_MFP_ENABLED) {
+			netdev_info(netdev,
+				    "Configuring TC not supported in MFP mode\n");
+			goto exit;
+		}
 
-	/* Check whether tc count is within enabled limit */
-	if (tc > i40e_pf_get_num_tc(pf)) {
-		netdev_info(netdev, "TC count greater than enabled on link for adapter\n");
-		goto exit;
+		/* Check whether tc count is within enabled limit */
+		if (num_tc > i40e_pf_get_num_tc(pf)) {
+			netdev_info(netdev,
+				    "TC count greater than enabled on link for adapter\n");
+			goto exit;
+		}
+		break;
+	case TC_MQPRIO_HW_OFFLOAD:
+		if (pf->flags & I40E_FLAG_DCB_ENABLED) {
+			netdev_info(netdev,
+				    "Full offload of TC Mqprio options is not supported when DCB is enabled\n");
+			goto exit;
+		}
+
+		/* Check if MFP enabled */
+		if (pf->flags & I40E_FLAG_MFP_ENABLED) {
Quoted 8 lines

drivers/net/ethernet/intel/i40e/i40e_main.c void i40e_down(struct i40e_vsi *vsi) · line 6582

Quoted 4 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +4 -0
+			netdev_info(netdev,
+				    "Configuring TC not supported in MFP mode\n");
+			goto exit;
+		}

This check is in both cases, and probably always needs to be checked, so it can probably be generalized and put once before this switch rather than repeating the code.

Right. Will fix in v3.

Quoted 57 lines
Quoted 42 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +18 -3
+		ret = i40e_validate_mqprio_qopt(vsi, mqprio_qopt);
+		if (ret)
+			goto exit;
+		memcpy(&vsi->mqprio_qopt, mqprio_qopt,
+		       sizeof(*mqprio_qopt));
+		pf->flags |= I40E_FLAG_TC_MQPRIO;
+		pf->flags &= ~I40E_FLAG_DCB_ENABLED;
+		break;
+	default:
+		return -EINVAL;
  	}
 
+config_tc:
  	/* Generate TC map for number of tc requested */
-	for (i = 0; i < tc; i++)
+	for (i = 0; i < num_tc; i++)
  		enabled_tc |= BIT(i);
 
  	/* Requesting same TC configuration as already enabled */
-	if (enabled_tc == vsi->tc_config.enabled_tc)
+	if (enabled_tc == vsi->tc_config.enabled_tc &&
+	    hw != TC_MQPRIO_HW_OFFLOAD)
  		return 0;
 
  	/* Quiesce VSI queues */
  	i40e_quiesce_vsi(vsi);
 
+	if (!hw && !(pf->flags & I40E_FLAG_TC_MQPRIO))
+		i40e_remove_queue_channels(vsi);
+
  	/* Configure VSI for enabled TCs */
  	ret = i40e_vsi_config_tc(vsi, enabled_tc);
  	if (ret) {
  		netdev_info(netdev, "Failed configuring TC for VSI seid=%d\n",
  			    vsi->seid);
-		goto exit;
+		goto reset;
  	}
 
  	if (pf->flags & I40E_FLAG_TC_MQPRIO) {

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_setup_tc(struct net_device *netdev, u8 tc) · line 6649

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +2 -2
@@ -6307,27 +6642,31 @@ static int i40e_setup_tc(struct net_device *netdev, u8 tc)
  		if (ret) {
  			netdev_info(netdev,
  				    "Failed configuring queue channels\n");
-			goto exit;
+			goto reset;
  		}
  	}
 
-exit:
+	goto out;

This "goto out" just to skip the reset call seems rather ugly. There's got to be a better way to do this and clean up the spaghetti, for the error cases and for the normal case.

Quoted 4 lines

If there is no need to do cleanup, all the "goto exit" calls could be "return ret", which gets rid of the "exit" label.

Maybe the need for unquiesce can be tracked by a bool?

Ok, I'll make this better in v3.

Quoted 130 lines
Quoted 130 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +11 -86
+reset:
+	/* Reset the configuration data to defaults, only TC0 is enabled */
+	i40e_vsi_set_default_tc_config(vsi);
+out:
  	/* Unquiesce VSI */
  	i40e_unquiesce_vsi(vsi);
+exit:
  	return ret;
  }
 
  static int __i40e_setup_tc(struct net_device *netdev, enum tc_setup_type type,
  			   void *type_data)
  {
-	struct tc_mqprio_qopt *mqprio = type_data;
-
-	if (type != TC_SETUP_MQPRIO)
+	switch (type) {
+	case TC_SETUP_MQPRIO:
+	case TC_SETUP_MQPRIO_EXT:
+		return i40e_setup_tc(netdev, type, type_data);
+	default:
  		return -EOPNOTSUPP;
-
-	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
-
-	return i40e_setup_tc(netdev, mqprio->num_tc);
+	}
  }
 
  /**
@@ -9174,45 +9513,6 @@ static int i40e_setup_misc_vector(struct i40e_pf *pf)
  }
 
  /**
- * i40e_config_rss_aq - Prepare for RSS using AQ commands
- * @vsi: vsi structure
- * @seed: RSS hash seed
- **/
-static int i40e_config_rss_aq(struct i40e_vsi *vsi, const u8 *seed,
-			      u8 *lut, u16 lut_size)
-{
-	struct i40e_pf *pf = vsi->back;
-	struct i40e_hw *hw = &pf->hw;
-	int ret = 0;
-
-	if (seed) {
-		struct i40e_aqc_get_set_rss_key_data *seed_dw =
-			(struct i40e_aqc_get_set_rss_key_data *)seed;
-		ret = i40e_aq_set_rss_key(hw, vsi->id, seed_dw);
-		if (ret) {
-			dev_info(&pf->pdev->dev,
-				 "Cannot set RSS key, err %s aq_err %s\n",
-				 i40e_stat_str(hw, ret),
-				 i40e_aq_str(hw, hw->aq.asq_last_status));
-			return ret;
-		}
-	}
-	if (lut) {
-		bool pf_lut = vsi->type == I40E_VSI_MAIN ? true : false;
-
-		ret = i40e_aq_set_rss_lut(hw, vsi->id, pf_lut, lut, lut_size);
-		if (ret) {
-			dev_info(&pf->pdev->dev,
-				 "Cannot set RSS lut, err %s aq_err %s\n",
-				 i40e_stat_str(hw, ret),
-				 i40e_aq_str(hw, hw->aq.asq_last_status));
-			return ret;
-		}
-	}
-	return ret;
-}
-
-/**
   * i40e_get_rss_aq - Get RSS keys and lut by using AQ commands
   * @vsi: Pointer to vsi structure
   * @seed: Buffter to store the hash keys
@@ -9259,46 +9559,6 @@ static int i40e_get_rss_aq(struct i40e_vsi *vsi, const u8 *seed,
  }
 
  /**
- * i40e_vsi_config_rss - Prepare for VSI(VMDq) RSS if used
- * @vsi: VSI structure
- **/
-static int i40e_vsi_config_rss(struct i40e_vsi *vsi)
-{
-	u8 seed[I40E_HKEY_ARRAY_SIZE];
-	struct i40e_pf *pf = vsi->back;
-	u8 *lut;
-	int ret;
-
-	if (!(pf->hw_features & I40E_HW_RSS_AQ_CAPABLE))
-		return 0;
-
-	if (!vsi->rss_size)
-		vsi->rss_size = min_t(int, pf->alloc_rss_size,
-				      vsi->num_queue_pairs);
-	if (!vsi->rss_size)
-		return -EINVAL;
-
-	lut = kzalloc(vsi->rss_table_size, GFP_KERNEL);
-	if (!lut)
-		return -ENOMEM;
-	/* Use the user configured hash keys and lookup table if there is one,
-	 * otherwise use default
-	 */
-	if (vsi->rss_lut_user)
-		memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size);
-	else
-		i40e_fill_rss_lut(pf, lut, vsi->rss_table_size, vsi->rss_size);
-	if (vsi->rss_hkey_user)
-		memcpy(seed, vsi->rss_hkey_user, I40E_HKEY_ARRAY_SIZE);
-	else
-		netdev_rss_key_fill((void *)seed, I40E_HKEY_ARRAY_SIZE);
-	ret = i40e_config_rss_aq(vsi, seed, lut, vsi->rss_table_size);
-	kfree(lut);
-
-	return ret;
-}
-
-/**
   * i40e_config_rss_reg - Configure RSS keys and lut by writing registers
   * @vsi: Pointer to vsi structure
   * @seed: RSS hash seed

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

#11 lore raw

PATCH v2 5/6 net: i40e: Refactor VF BW rate limiting function to be reused on the PF as well

In reply to Amritha Nambiar

This patch refactors the BW rate limiting for Tx traffic on the VF to be reused in the next patch for rate limiting Tx traffic for the VSIs on the PF as well.

---
 drivers/net/ethernet/intel/i40e/i40e.h             |    5 ++
 drivers/net/ethernet/intel/i40e/i40e_main.c        |   63 ++++++++++++++++++++
 drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c |   45 +-------------
 3 files changed, 70 insertions(+), 43 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +5 -0
index 2565280..304921e 100644
@@ -128,6 +128,10 @@
 /* default to trying for four seconds */
 #define I40E_TRY_LINK_TIMEOUT	(4 * HZ)
 
+/* BW rate limiting */
+#define I40E_BW_CREDIT_DIVISOR		50 /* 50Mbps per BW credit */
+#define I40E_MAX_BW_INACTIVE_ACCUM	4  /* accumulate 4 credits max */
+
 /* driver state flags */
 enum i40e_state_t {
 	__I40E_TESTING,
@@ -1042,4 +1046,5 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi)
 }
 
 int i40e_create_queue_channel(struct i40e_vsi *vsi, struct i40e_channel *ch);
+int i40e_set_bw_limit(struct i40e_vsi *vsi, u16 seid, u64 max_tx_rate);
 #endif /* _I40E_H_ */
drivers/net/ethernet/intel/i40e/i40e_main.c +63 -0

69 lines not on this page · raw message

drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +2 -43

65 lines not on this page · raw message

#12 lore raw

PATCH v2 5/6 net: i40e: Refactor VF BW rate limiting function to be reused on the PF as well

Quoted 76 lines · Amritha Nambiar

This patch refactors the BW rate limiting for Tx traffic on the VF to be reused in the next patch for rate limiting Tx traffic for the VSIs on the PF as well.

---
  drivers/net/ethernet/intel/i40e/i40e.h             |    5 ++
  drivers/net/ethernet/intel/i40e/i40e_main.c        |   63 ++++++++++++++++++++
  drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c |   45 +-------------
  3 files changed, 70 insertions(+), 43 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +5 -0
index 2565280..304921e 100644
@@ -128,6 +128,10 @@
  /* default to trying for four seconds */
  #define I40E_TRY_LINK_TIMEOUT	(4 * HZ)
 
+/* BW rate limiting */
+#define I40E_BW_CREDIT_DIVISOR		50 /* 50Mbps per BW credit */
+#define I40E_MAX_BW_INACTIVE_ACCUM	4  /* accumulate 4 credits max */
+
  /* driver state flags */
  enum i40e_state_t {
  	__I40E_TESTING,
@@ -1042,4 +1046,5 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi)
  }
 
  int i40e_create_queue_channel(struct i40e_vsi *vsi, struct i40e_channel *ch);
+int i40e_set_bw_limit(struct i40e_vsi *vsi, u16 seid, u64 max_tx_rate);
  #endif /* _I40E_H_ */
drivers/net/ethernet/intel/i40e/i40e_main.c +36 -0
index 5b783fc..fcf563a 100644
@@ -5405,6 +5405,69 @@ static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc)
  }
 
  /**
+ * i40e_get_link_speed - Returns link speed for the interface
+ * @vsi: VSI to be configured
+ *
+ **/
+int i40e_get_link_speed(struct i40e_vsi *vsi)
+{
+	struct i40e_pf *pf = vsi->back;
+
+	switch (pf->hw.phy.link_info.link_speed) {
+	case I40E_LINK_SPEED_40GB:
+		return 40000;
+	case I40E_LINK_SPEED_25GB:
+		return 25000;
+	case I40E_LINK_SPEED_20GB:
+		return 20000;
+	case I40E_LINK_SPEED_10GB:
+		return 10000;
+	case I40E_LINK_SPEED_1GB:
+		return 1000;
+	default:
+		return -EINVAL;
+	}
+}
+
+/**
+ * i40e_set_bw_limit - setup BW limit for Tx traffic based on max_tx_rate
+ * @vsi: VSI to be configured
+ * @seid: seid of the channel/VSI
+ * @max_tx_rate: max TX rate to be configured as BW limit
+ *
+ * Helper function to set BW limit for a given VSI
+ **/
+int i40e_set_bw_limit(struct i40e_vsi *vsi, u16 seid, u64 max_tx_rate)
+{
+	struct i40e_pf *pf = vsi->back;
+	int speed = 0;

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5453

Quoted 10 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +10 -0
+	int ret = 0;
+
+	speed = i40e_get_link_speed(vsi);
+	if (max_tx_rate > speed) {
+		dev_err(&pf->pdev->dev,
+			"Invalid max tx rate %llu specified for VSI seid %d.",
+			max_tx_rate, seid);
+		return -EINVAL;
+	}
+	if ((max_tx_rate < 50) && (max_tx_rate > 0)) {

if (max_tx_rate && (max_tx_rate < 50)) {

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5459

Quoted 6 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +6 -0
+		dev_warn(&pf->pdev->dev,
+			 "Setting max tx rate to minimum usable value of 50Mbps.\n");
+		max_tx_rate = 50;
+	}
+
+       /* Tx rate credits are in values of 50Mbps, 0 is disabled*/

This should be indented with a tab, not spaces

drivers/net/ethernet/intel/i40e/i40e_main.c static int i40e_vsi_config_tc(struct i40e_vsi *vsi, u8 enabled_tc) · line 5466

Quoted 7 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +7 -0
+	ret = i40e_aq_config_vsi_bw_limit(&pf->hw, seid,
+					  max_tx_rate / I40E_BW_CREDIT_DIVISOR,
+					  I40E_MAX_BW_INACTIVE_ACCUM, NULL);
+	if (ret)
+		dev_err(&pf->pdev->dev,
+			"Failed set tx rate (%llu Mbps) for vsi->seid %u, error code %d.\n",
+			max_tx_rate, seid, ret);

Please do the AQ error decode here with i40e_stat_str() and i40e_aq_str()

Quoted 85 lines
drivers/net/ethernet/intel/i40e/i40e_main.c +4 -0
+	return ret;
+}
+
+/**
   * i40e_remove_queue_channels - Remove queue channels for the TCs
   * @vsi: VSI to be configured
   *
drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +2 -43
index 92fe06d..a1d2391 100644
@@ -3061,8 +3061,6 @@ int i40e_ndo_set_vf_port_vlan(struct net_device *netdev, int vf_id,
  	return ret;
  }
 
-#define I40E_BW_CREDIT_DIVISOR 50     /* 50Mbps per BW credit */
-#define I40E_MAX_BW_INACTIVE_ACCUM 4  /* device can accumulate 4 credits max */
  /**
   * i40e_ndo_set_vf_bw
   * @netdev: network interface device structure
@@ -3078,7 +3076,6 @@ int i40e_ndo_set_vf_bw(struct net_device *netdev, int vf_id, int min_tx_rate,
  	struct i40e_pf *pf = np->vsi->back;
  	struct i40e_vsi *vsi;
  	struct i40e_vf *vf;
-	int speed = 0;
  	int ret = 0;
 
  	/* validate the request */
@@ -3103,48 +3100,10 @@ int i40e_ndo_set_vf_bw(struct net_device *netdev, int vf_id, int min_tx_rate,
  		goto error;
  	}
 
-	switch (pf->hw.phy.link_info.link_speed) {
-	case I40E_LINK_SPEED_40GB:
-		speed = 40000;
-		break;
-	case I40E_LINK_SPEED_25GB:
-		speed = 25000;
-		break;
-	case I40E_LINK_SPEED_20GB:
-		speed = 20000;
-		break;
-	case I40E_LINK_SPEED_10GB:
-		speed = 10000;
-		break;
-	case I40E_LINK_SPEED_1GB:
-		speed = 1000;
-		break;
-	default:
-		break;
-	}
-
-	if (max_tx_rate > speed) {
-		dev_err(&pf->pdev->dev, "Invalid max tx rate %d specified for VF %d.\n",
-			max_tx_rate, vf->vf_id);
-		ret = -EINVAL;
+	ret = i40e_set_bw_limit(vsi, vsi->seid, max_tx_rate);
+	if (ret)
  		goto error;
-	}
 
-	if ((max_tx_rate < 50) && (max_tx_rate > 0)) {
-		dev_warn(&pf->pdev->dev, "Setting max Tx rate to minimum usable value of 50Mbps.\n");
-		max_tx_rate = 50;
-	}
-
-	/* Tx rate credits are in values of 50Mbps, 0 is disabled*/
-	ret = i40e_aq_config_vsi_bw_limit(&pf->hw, vsi->seid,
-					  max_tx_rate / I40E_BW_CREDIT_DIVISOR,
-					  I40E_MAX_BW_INACTIVE_ACCUM, NULL);
-	if (ret) {
-		dev_err(&pf->pdev->dev, "Unable to set max tx rate, error code %d.\n",
-			ret);
-		ret = -EIO;
-		goto error;
-	}
  	vf->tx_rate = max_tx_rate;
  error:
  	return ret;

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan at osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

#13 lore raw

PATCH v2 6/6 net: i40e: Add support to set max bandwidth rates for TCs offloaded via tc/mqprio

In reply to Amritha Nambiar

This patch enables setting up maximum Tx rates for the traffic classes in i40e. The maximum rate offloaded to the hardware through the mqprio framework is configured for the VSI. Configuring minimum Tx rate limit is not supported in the device. The minimum usable value for Tx rate is 50Mbps.

Example:

# tc qdisc add dev eth0 root mqprio num_tc 2  map 0 0 0 0 1 1 1 1\
  queues 4 at 0 4 at 4 max_rate 55Mbit 60Mbit hw 2

To dump the bandwidth rates:

# tc qdisc show dev eth0

qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0

             queues:(0:3) (4:7)
             min_rate:0bit 0bit
             max_rate:55Mbit 60Mbit
---
 drivers/net/ethernet/intel/i40e/i40e.h      |    2 +
 drivers/net/ethernet/intel/i40e/i40e_main.c |   99 +++++++++++++++++++++++++--
 2 files changed, 92 insertions(+), 9 deletions(-)
drivers/net/ethernet/intel/i40e/i40e.h +2 -0
index 304921e..8852ac0 100644
@@ -359,6 +359,8 @@ struct i40e_channel {
 	u8 enabled_tc;
 	struct i40e_aqc_vsi_properties_data info;
 
+	u64 max_tx_rate;
+
 	/* track this channel belongs to which VSI */
 	struct i40e_vsi *parent_vsi;
 };
drivers/net/ethernet/intel/i40e/i40e_main.c +90 -9

190 lines not on this page · raw message